SELinux System Administration: Implement mandatory access control to secure applications, users, and information flows on Linux [3 ed.] 1800201478, 9781800201477

Table of contents :
Cover
Title Page
Copyright and Credits
About Packt
Contributors
Table of Contents
Preface
Section 1: Using SELinux
Chapter 1: Fundamental SELinux Concepts
Technical requirements
Providing more security for Linux
Introducing Linux Security Modules (LSM)
Extending regular DAC with SELinux
Restricting root privileges
Reducing the impact of vulnerabilities
Enabling SELinux support
Labeling all resources and objects
Dissecting the SELinux context
Enforcing access through types
Granting domain access through roles
Limiting roles through users
Controlling information flow through sensitivities
Defining and distributing policies
Writing SELinux policies
Distributing policies through modules
Bundling modules in a policy store
Distinguishing between policies
Supporting MLS
Dealing with unknown permissions
Supporting unconfined domains
Limiting cross-user sharing
Incrementing policy versions
Different policy content
Summary
Questions
Chapter 2: Understanding SELinux Decisions and Logging
Technical requirements
Switching SELinux on and off
Setting the global SELinux state
Switching to permissive or enforcing mode
Using kernel boot parameters
Disabling SELinux protections for a single service
Understanding SELinux-aware applications
SELinux logging and auditing
Following audit events
Tuning the AVC
Uncovering more logging
Configuring Linux auditing
Configuring the local system logger
Reading SELinux denials
Other SELinux-related event types
Using ausearch
Getting help with denials
Troubleshooting with setroubleshoot
Sending emails when SELinux denials occur
Using audit2why
Interacting with systemd-journal
Using common sense
Summary
Questions
Chapter 3: Managing User Logins
Technical requirements
User-oriented SELinux contexts
SELinux users and roles
Listing SELinux user mappings
Mapping logins to SELinux users
Customizing logins for services
Creating SELinux users
Listing accessible domains
Managing categories
Handling SELinux roles
Defining allowed SELinux contexts
Validating contexts with getseuser
Switching roles with newrole
Managing role access through sudo
Reaching other domains using runcon
Switching to the system role
SELinux and PAM
Assigning contexts through PAM
Prohibiting access during permissive mode
Polyinstantiating directories
Summary
Questions
Chapter 4: Using File Contexts and Process Domains
Technical requirements
Introduction to SELinux file contexts
Getting context information
Interpreting SELinux context types
Keeping or ignoring contexts
Inheriting the default contexts
Querying transition rules
Copying and moving files
Temporarily changing file contexts
Placing categories on files and directories
Using multilevel security on files
Backing up and restoring extended attributes
Using mount options to set SELinux contexts
SELinux file context expressions
Using context expressions
Registering file context changes
Optimizing recursive context operations
Using customizable types
Compiling the different file_contexts files
Exchanging local modifications
Modifying file contexts
Using setfiles, rlpkg, and fixfiles
Relabeling the entire filesystem
Automatically setting context with restorecond
Setting SELinux context at boot with tmpfiles
The context of a process
Getting a process context
Transitioning toward a domain
Verifying a target context
Other supported transitions
Querying initial contexts
Tweaking memory protections
Limiting the scope of transitions
Sanitizing environments on transition
Disabling unconstrained transitions
Using Linux's NO_NEW_PRIVS
Types, permissions, and constraints
Understanding type attributes
Querying domain permissions
Learning about constraints
Summary
Questions
Chapter 5: Controlling Network Communications
Technical requirements
Controlling process communications
Using shared memory
Communicating locally through pipes
Conversing over UNIX domain sockets
Understanding netlink sockets
Dealing with TCP, UDP, and SCTP sockets
Listing connection contexts
Linux firewalling and SECMARK support
Introducing netfilter
Implementing security markings
Assigning labels to packets
Transitioning to nftables
Assessing eBPF
Securing high-speed InfiniBand networks
Directly accessing memory
Protecting InfiniBand networks
Managing the InfiniBand subnet
Controlling access to InfiniBand partitions
Understanding labeled networking
Fallback labeling with NetLabel
Limiting flows based on the network interface
Accepting peer communication from selected hosts
Verifying peer-to-peer flow
Using old-style controls
Using labeled IPsec with SELinux
Setting up regular IPsec
Enabling labeled IPsec
Supporting CIPSO with NetLabel and SELinux
Configuring CIPSO mappings
Adding domain-specific mappings
Using local CIPSO definitions
Supporting IPv6 CALIPSO
Summary
Questions
Chapter 6: Configuring SELinux through Infrastructure-as-Code Orchestration
Technical requirements
Introducing the target settings and policies
The idempotency of actions
Policy and state management
SELinux configuration settings
Setting file contexts
Recovering from mistakes
Comparing frameworks
Using Ansible for SELinux system administration
How Ansible works
Installing and configuring Ansible
Creating and testing the Ansible role
Assigning SELinux contexts to filesystem resources with Ansible
Loading custom SELinux policies with Ansible
Using Ansible's out-of-the-box SELinux support
Utilizing SaltStack to configure SELinux
How SaltStack works
Installing and configuring SaltStack
Creating and testing our SELinux state with SaltStack
Assigning SELinux contexts to filesystem resources with SaltStack
Loading custom SELinux policies with SaltStack
Using SaltStack's out-of-the-box SELinux support
Automating system management with Puppet
How Puppet works
Installing and configuring Puppet
Creating and testing the SELinux class with Puppet
Assigning SELinux contexts to filesystem resources with Puppet
Loading custom SELinux policies with Puppet
Using Puppet's out-of-the-box SELinux support
Wielding Chef for system automation
How Chef works
Installing and configuring Chef
Creating the SELinux cookbook
Assigning SELinux contexts to filesystem resources with Chef
Loading custom SELinux policies with Chef
Using Chef's out-of-the-box SELinux support
Summary
Questions
Section 2: SELinux-Aware Platforms
Chapter 7: Configuring Application-Specific SELinux Controls
Technical requirements
Tuning systemd services, logging, and device management
Service support in systemd
Logging with systemd
Handling device files
Communicating over D-Bus
Understanding D-Bus
Controlling service acquisition with SELinux
Governing message flows
Configuring PAM services
Cockpit
Cron
OpenSSH
Using mod_selinux with Apache
Introducing mod_selinux
Configuring the general Apache SELinux sensitivity
Mapping end users to specific domains
Changing domains based on source
Summary
Questions
Chapter 8: SEPostgreSQL – Extending PostgreSQL with SELinux
Technical requirements
Introducing PostgreSQL and sepgsql
Reconfiguring PostgreSQL with sepgsql
Creating a test account
Tuning sepgsql inside PostgreSQL
Troubleshooting sepgsql
Understanding SELinux's database-specific object classes and permissions
Understanding sepgsql permissions
Using the default supported types
Creating trusted procedures
Using sepgsql-specific functions
Using MCS and MLS
Limiting access to columns based on categories
Constraining the user domain for sensitivity range manipulation
Integrating SEPostgreSQL into the network
Creating a fallback label for remote sessions
Tuning the SELinux policy
Summary
Questions
Chapter 9: Secure Virtualization
Technical requirements
Understanding SELinux-secured virtualization
Introducing virtualization
Reviewing the risks of virtualization
Reusing existing virtualization domains
Fine-tuning virtualization-supporting SELinux policy
Understanding sVirt's use of MCS
Enhancing libvirt with SELinux support
Differentiating between shared and dedicated resources
Assessing the libvirt architecture
Configuring libvirt for sVirt
Changing a guest's SELinux labels
Customizing resource labels
Controlling available categories
Changing the storage pool locations
Using Vagrant with libvirt
Deploying Vagrant and the libvirt plugin
Installing a libvirt-compatible box
Configuring Vagrant boxes
Summary
Questions
Chapter 10: Using Xen Security Modules with FLASK
Technical requirements
Understanding Xen and XSM
Introducing the Xen hypervisor
Installing Xen
Creating an unprivileged guest
Understanding Xen Security Modules
Running XSM-enabled Xen
Rebuilding Xen with XSM support
Using XSM labels
Manipulating XSM
Applying custom XSM policies
Summary
Questions
Chapter 11: Enhancing the Security of Containerized Workloads
Technical requirements
Using SELinux with systemd's container support
Initializing a systemd container
Using a specific SELinux context
Facilitating container management with machinectl
Configuring podman
Selecting podman over Docker
Using containers with SELinux
Changing a container's SELinux domain
Creating custom domains with udica
Toggling container_t privileges with SELinux booleans
Tuning the container hosting environment
Leveraging Kubernetes' SELinux support
Configuring Kubernetes with SELinux support
Setting SELinux contexts for pods
Summary
Questions
Section 3: Policy Management
Chapter 12: Tuning SELinux Policies
Technical requirements
Working with SELinux booleans
Listing SELinux booleans
Changing boolean values
Inspecting the impact of a boolean
Handling policy modules
Listing policy modules
Loading and removing policy modules
Replacing and updating existing policies
Creating policies using audit2allow
Using sensible module names
Generating reference policy style modules with audit2allow
Building reference policy - style modules
Building legacy-style modules
Replacing the default distribution policy
Summary
Questions
Chapter 13: Analyzing Policy Behavior
Technical requirements
Performing single-step analysis
Using different SELinux policy files
Displaying policy object information
Understanding sesearch
Querying allow rules
Querying type transition rules
Querying other type rules
Querying role-related rules
Browsing with apol
Using apol workspaces
Investigating domain transitions
Using apol for domain transition analysis
Using sedta for domain transition analysis
Using sepolicy for domain transition analysis
Analyzing information flow
Using apol for information flow analysis
Using seinfoflow for information flow analysis
Using sepolicy communicate for simple information flow analysis
Comparing policies
Using sediff to compare policies
Summary
Questions
Chapter 14: Dealing with New Applications
Technical requirements
Running applications without restrictions
Understanding how unconfined domains work
Making new applications run as an unconfined domain
Extending unconfined domains
Marking domains as permissive
Using sandboxed applications
Understanding the SELinux sandbox
Using the sandbox command
Assigning common policies to new applications
Understanding domain complexity
Running applications in a specific policy
Extending generated policies
Understanding the limitations of generated policies
Introducing sepolicy generate
Generating policies with sepolicy generate
Summary
Questions
Chapter 15: Using the Reference Policy
Technical requirements
Introducing the reference policy
Navigating the policy
Structuring policy modules
Using and understanding the policy macros
Making use of single-class permission groups
Calling permission groups
Creating application-level policies
Constructing network-facing service policies
Addressing user applications
Adding user-level policies
Getting help with supporting tools
Verifying code with selint
Querying the interfaces and macros locally
Summary
Questions
Chapter 16: Developing Policies with SELinux CIL
Technical requirements
Introducing CIL
Translating .pp files to CIL
Understanding CIL syntax
Creating fine-grained definitions
Depending on roles or types
Defining a new port type
Adding constraints to the policy
Building complete application policies
Using namespaces
Extending the policy with attribute assignments
Adding entry point information
Gradually extending the policy further
Introducing permission sets
Adding macros
Summary
Questions
Assessments
Other Books You May Enjoy
Index

Citation preview

---

SELinux System Administration Third Edition Implement mandatory access control to secure applications, users, and information flows on Linux

Sven Vermeulen

BIRMINGHAM—MUMBAI

SELinux System Administration Third Edition Copyright © 2020 Packt Publishing All rights reserved. No part of this book may be reproduced, stored in a retrieval system, or transmitted in any form or by any means, without the prior written permission of the publisher, except in the case of brief quotations embedded in critical articles or reviews. Every effort has been made in the preparation of this book to ensure the accuracy of the information presented. However, the information contained in this book is sold without warranty, either express or implied. Neither the author, nor Packt Publishing or its dealers and distributors, will be held liable for any damages caused or alleged to have been caused directly or indirectly by this book. Packt Publishing has endeavored to provide trademark information about all of the companies and products mentioned in this book by the appropriate use of capitals. However, Packt Publishing cannot guarantee the accuracy of this information. Commissioning Editor: Vijin Boricha Acquisition Editor: Shrilekha Inani Senior Editor: Arun Nadar Content Development Editor: Romy Dias Technical Editor: Soham Amburle Copy Editor: Safis Editing Project Coordinator: Neil Dmello Proofreader: Safis Editing Indexer: Pratik Shirodkar Production Designer: Vijay Kamble First published: September 2013 Second edition: December 2016 Third edition: November 2020 Production reference: 1041120 Published by Packt Publishing Ltd. Livery Place 35 Livery Street Birmingham B3 2PB, UK. ISBN 978-1-80020-147-7

www.packt.com

To the doctors, nurses, public health officials, and first responders who are protecting us from COVID-19.

Packt.com

Subscribe to our online digital library for full access to over 7,000 books and videos, as well as industry leading tools to help you plan your personal development and advance your career. For more information, please visit our website.

Why subscribe? • Spend less time learning and more time coding with practical eBooks and Videos from over 4,000 industry professionals • Improve your learning with Skill Plans built especially for you • Get a free eBook or video every month • Fully searchable for easy access to vital information • Copy and paste, print, and bookmark content Did you know that Packt offers eBook versions of every book published, with PDF and ePub files available? You can upgrade to the eBook version at packt.com and as a print book customer, you are entitled to a discount on the eBook copy. Get in touch with us at [email protected] for more details. At www.packt.com, you can also read a collection of free technical articles, sign up for a range of free newsletters, and receive exclusive discounts and offers on Packt books and eBooks.

Contributors About the author Sven Vermeulen (sjvermeu on Twitter) is a long-term contributor to various free software projects and the author of several online guides and resources, including the Gentoo Handbook. He got his first taste of free software in 1997 and never looked back. Within SELinux, Sven contributed several policies to the Reference Policy project, and actively participated in policy development and user space development projects. In his daily job, Sven is an enterprise architect in a European financial institution as well as a self-employed solution engineer and consultant. Prior to this, he graduated with an MSE in computer engineering from Ghent University and an MSc in ICT enterprise architecture from IC Institute. I want to thank the SELinux community at large for being who they are, mature in their answers, and knowledgeable in all areas. Closer to my heart, I would like to thank my daughter for giving me the time to work on this book, despite being locked in with me during the COVID-19 lockdown measures.

About the reviewers David Windsor is a free software hacker who fell in love with Linux when he received his first Slackware installation disc at a Linux Users' Group in the late 1990s. He initially started contributing to Linux security as a hobbyist but has since been fortunate enough to be able to get paid to do what he loves. David currently works at FireEye, Inc., where he does operating system security research. Previously, David has worked at Red Hat and The Linux Foundation and has contributed code to the Kernel Self-Protection Project (KSPP). I'd like to thank Drew and Liz, loyal party members who have been there from the start. Rock on London, rock on Chicago!

Jimmy Savage is a certified Linux system administrator, DevOps tools engineer, network engineer, and security researcher. He focuses on supporting mission-critical devices and infrastructure, including everything from flight simulators, embedded systems, data centers, and beyond.

Packt is searching for authors like you If you're interested in becoming an author for Packt, please visit authors. packtpub.com and apply today. We have worked with thousands of developers and tech professionals, just like you, to help them share their insight with the global tech community. You can make a general application, apply for a specific hot topic that we are recruiting an author for, or submit your own idea.

Table of Contents Preface

Section 1: Using SELinux

1

Fundamental SELinux Concepts Technical requirements Providing more security for Linux

20

Defining and distributing policies36

20

Writing SELinux policies Distributing policies through modules Bundling modules in a policy store

Introducing Linux Security Modules (LSM)22 Extending regular DAC with SELinux 25 Restricting root privileges 26 Reducing the impact of vulnerabilities 26 Enabling SELinux support 27

Labeling all resources and objects

28

Dissecting the SELinux context 30 Enforcing access through types 31 Granting domain access through roles 32 Limiting roles through users 34 Controlling information flow through sensitivities35

37 39 40

Distinguishing between policies41 Supporting MLS Dealing with unknown permissions Supporting unconfined domains Limiting cross-user sharing Incrementing policy versions Different policy content

41 41 42 43 44 45

Summary47 Questions47

ii Table of Contents

2

Understanding SELinux Decisions and Logging Technical requirements Switching SELinux on and off

50 50

Setting the global SELinux state 50 Switching to permissive or enforcing mode 52 Using kernel boot parameters 54 Disabling SELinux protections for a single service 55 Understanding SELinux-aware applications56

SELinux logging and auditing

57

Following audit events Tuning the AVC Uncovering more logging

58 60 61

Configuring Linux auditing Configuring the local system logger Reading SELinux denials Other SELinux-related event types Using ausearch

62 64 65 68 71

Getting help with denials

72

Troubleshooting with setroubleshoot Sending emails when SELinux denials occur Using audit2why Interacting with systemd-journal Using common sense

73 74 75 76 77

Summary79 Questions79

3

Managing User Logins Technical requirements 82 User-oriented SELinux contexts 82 SELinux users and roles 83 Listing SELinux user mappings Mapping logins to SELinux users Customizing logins for services Creating SELinux users Listing accessible domains Managing categories

83 84 86 87 89 90

Handling SELinux roles

91

Defining allowed SELinux contexts Validating contexts with getseuser

91 93

Switching roles with newrole Managing role access through sudo Reaching other domains using runcon Switching to the system role

93 94 95 96

SELinux and PAM

96

Assigning contexts through PAM 97 Prohibiting access during permissive mode98 Polyinstantiating directories 99

Summary101 Questions101

Table of Contents iii

4

Using File Contexts and Process Domains Technical requirements Introduction to SELinux file contexts

104 104

Getting context information Interpreting SELinux context types

105 107

Keeping or ignoring contexts

108

Inheriting the default contexts Querying transition rules Copying and moving files Temporarily changing file contexts Placing categories on files and directories Using multilevel security on files Backing up and restoring extended attributes Using mount options to set SELinux contexts

109 109 110 112 113 113 114 114

SELinux file context expressions115 Using context expressions 116 Registering file context changes 117 Optimizing recursive context operations119 Using customizable types 120 Compiling the different file_contexts files 121 Exchanging local modifications 122

Modifying file contexts

122

Using setfiles, rlpkg, and fixfiles Relabeling the entire filesystem Automatically setting context with restorecond Setting SELinux context at boot with tmpfiles

123 123 124 124

The context of a process

125

Getting a process context Transitioning toward a domain Verifying a target context Other supported transitions Querying initial contexts Tweaking memory protections

126 127 129 130 130 130

Limiting the scope of transitions

131

Sanitizing environments on transition 132 Disabling unconstrained transitions132 Using Linux's NO_NEW_PRIVS 134

Types, permissions, and constraints136 Understanding type attributes Querying domain permissions Learning about constraints

136 137 138

Summary139 Questions140

iv Table of Contents

5

Controlling Network Communications Technical requirements 142 Controlling process communications142 Using shared memory Communicating locally through pipes Conversing over UNIX domain sockets Understanding netlink sockets Dealing with TCP, UDP, and SCTP sockets Listing connection contexts

Linux firewalling and SECMARK support Introducing netfilter Implementing security markings Assigning labels to packets Transitioning to nftables Assessing eBPF

Securing high-speed InfiniBand networks Directly accessing memory Protecting InfiniBand networks Managing the InfiniBand subnet Controlling access to InfiniBand

142 143 144 145 145 147

148 148 150 151 153 154

157 158 158 159

partitions161

Understanding labeled networking163 Fallback labeling with NetLabel Limiting flows based on the network interface Accepting peer communication from selected hosts Verifying peer-to-peer flow Using old-style controls

163 164 165 165 166

Using labeled IPsec with SELinux167 Setting up regular IPsec Enabling labeled IPsec

Supporting CIPSO with NetLabel and SELinux Configuring CIPSO mappings Adding domain-specific mappings Using local CIPSO definitions Supporting IPv6 CALIPSO

169 169

170 171 172 173 173

Summary174 Questions174

6

Configuring SELinux through Infrastructure-as-Code Orchestration Technical requirements Introducing the target settings and policies The idempotency of actions

176 176 176

Policy and state management SELinux configuration settings Setting file contexts Recovering from mistakes Comparing frameworks

177 177 178 178 178

Table of Contents v

Using Ansible for SELinux system administration How Ansible works Installing and configuring Ansible Creating and testing the Ansible role Assigning SELinux contexts to filesystem resources with Ansible Loading custom SELinux policies with Ansible Using Ansible's out-of-the-box SELinux support

Utilizing SaltStack to configure SELinux How SaltStack works Installing and configuring SaltStack Creating and testing our SELinux state with SaltStack Assigning SELinux contexts to filesystem resources with SaltStack Loading custom SELinux policies with SaltStack Using SaltStack's out-of-the-box SELinux support

179 179 180 181 182 183 185

186 186 187 188 189 190 190

Automating system

management with Puppet How Puppet works Installing and configuring Puppet Creating and testing the SELinux class with Puppet Assigning SELinux contexts to filesystem resources with Puppet Loading custom SELinux policies with Puppet Using Puppet's out-of-the-box SELinux support

191 191 192 193 194 195 196

Wielding Chef for system automation197 How Chef works Installing and configuring Chef Creating the SELinux cookbook Assigning SELinux contexts to filesystem resources with Chef Loading custom SELinux policies with Chef Using Chef's out-of-the-box SELinux support

197 198 201 203 203 204

Summary205 Questions206

Section 2: SELinux-Aware Platforms

7

Configuring Application-Specific SELinux Controls Technical requirements 210 Tuning systemd services, logging, and device management210 Service support in systemd Logging with systemd

212 220

Handling device files

Communicating over D-Bus Understanding D-Bus Controlling service acquisition with SELinux Governing message flows

221

224 224 226 227

vi Table of Contents

Configuring PAM services

228

Cockpit228 Cron230 OpenSSH232

Using mod_selinux with Apache Introducing mod_selinux

236 236

Configuring the general Apache SELinux sensitivity Mapping end users to specific domains Changing domains based on source

238 239 239

Summary241 Questions241

8

SEPostgreSQL – Extending PostgreSQL with SELinux Technical requirements Introducing PostgreSQL and sepgsql Reconfiguring PostgreSQL with sepgsql Creating a test account Tuning sepgsql inside PostgreSQL Troubleshooting sepgsql

Understanding SELinux's database-specific object classes and permissions Understanding sepgsql permissions Using the default supported types Creating trusted procedures

244 244 244 246 248 249

250 251 252 255

Using sepgsql-specific functions

Using MCS and MLS Limiting access to columns based on categories Constraining the user domain for sensitivity range manipulation

Integrating SEPostgreSQL into the network Creating a fallback label for remote sessions Tuning the SELinux policy

257

257 258 259

259 260 261

Summary261 Questions262

9

Secure Virtualization Technical requirements 263 Understanding SELinux-secured virtualization 264

Fine-tuning virtualization-supporting SELinux policy Understanding sVirt's use of MCS

Introducing virtualization 264 Reviewing the risks of virtualization 266 Reusing existing virtualization domains267

Enhancing libvirt with SELinux support Differentiating between shared and dedicated resources

268 270

271 272

Table of Contents vii Assessing the libvirt architecture Configuring libvirt for sVirt Changing a guest's SELinux labels Customizing resource labels Controlling available categories Changing the storage pool locations

Using Vagrant with libvirt

272 273 276 278 278 279

281

Deploying Vagrant and the libvirt plugin Installing a libvirt-compatible box Configuring Vagrant boxes

281 282 283

Summary283 Questions284

10

Using Xen Security Modules with FLASK Technical requirements Understanding Xen and XSM Introducing the Xen hypervisor Installing Xen Creating an unprivileged guest Understanding Xen Security Modules

Running XSM-enabled Xen

286 286 286 287 290 292

293

Rebuilding Xen with XSM support Using XSM labels Manipulating XSM

293 295 296

Applying custom XSM policies 299 Summary300 Questions300

11

Enhancing the Security of Containerized Workloads Technical requirements Using SELinux with systemd's container support

302

Initializing a systemd container Using a specific SELinux context Facilitating container management with machinectl

303 304

Configuring podman Selecting podman over Docker Using containers with SELinux Changing a container's SELinux domain

302

305

306 306 308 310

Creating custom domains with udica 311 Toggling container_t privileges with SELinux booleans 312 Tuning the container hosting environment312

Leveraging Kubernetes' SELinux support Configuring Kubernetes with SELinux support Setting SELinux contexts for pods

313 313 315

Summary317 Questions317

viii Table of Contents

Section 3: Policy Management

12

Tuning SELinux Policies Technical requirements 322 Working with SELinux booleans322 Listing SELinux booleans Changing boolean values Inspecting the impact of a boolean

Handling policy modules

322 323 325

325

Listing policy modules 326 Loading and removing policy modules 327

Replacing and updating existing policies

Creating policies using audit2allow 328 Using sensible module names 330 Generating reference policy style modules with audit2allow 331 Building reference policy - style modules332 Building legacy-style modules 332 Replacing the default distribution policy 333

Summary334 Questions334

328

13

Analyzing Policy Behavior Technical requirements 336 Performing single-step analysis336 Using different SELinux policy files Displaying policy object information Understanding sesearch Querying allow rules Querying type transition rules Querying other type rules Querying role-related rules Browsing with apol Using apol workspaces

336 337 340 341 342 343 344 345 348

Investigating domain transitions349

Using apol for domain transition analysis Using sedta for domain transition analysis Using sepolicy for domain transition analysis

Analyzing information flow Using apol for information flow analysis Using seinfoflow for information flow analysis Using sepolicy communicate for simple information flow analysis

Comparing policies

350 351 352

354 355 357 357

358

Table of Contents ix Using sediff to compare policies358

Summary359 Questions359

14

Dealing with New Applications Technical requirements Running applications without restrictions

362

Assigning common policies to new applications

362

Understanding domain complexity Running applications in a specific policy

372

Extending generated policies

375

Understanding how unconfined domains work Making new applications run as an unconfined domain Extending unconfined domains Marking domains as permissive

362 363 367 368

Using sandboxed applications 369 Understanding the SELinux sandbox Using the sandbox command

370 370

Understanding the limitations of generated policies Introducing sepolicy generate Generating policies with sepolicy generate

372

374

376 376 378

Summary379 Questions379

15

Using the Reference Policy Technical requirements 382 Introducing the reference policy382

Constructing network-facing service policies Addressing user applications

Navigating the policy Structuring policy modules

Adding user-level policies Getting help with supporting tools

Using and understanding the policy macros Making use of single-class permission groups Calling permission groups

383 384

390 390 391

Creating application-level policies392

Verifying code with selint Querying the interfaces and macros locally

392 396

398 399 399 400

Summary402 Questions402

x Table of Contents

16

Developing Policies with SELinux CIL Technical requirements Introducing CIL

404 404

Translating .pp files to CIL Understanding CIL syntax

404 405

Creating fine-grained definitions

406

Depending on roles or types Defining a new port type Adding constraints to the policy

406 407 408

Assessments Other Books You May Enjoy Index

Building complete application policies Using namespaces Extending the policy with attribute assignments Adding entry point information Gradually extending the policy further Introducing permission sets Adding macros

410 410 411 412 413 415 416

Summary418 Questions418

Preface Security-Enhanced Linux (SELinux) is one of the most complete security solutions for Linux and is available by default in most major Linux distributions, such as Red Hat Enterprise Linux, CentOS, Fedora, and Gentoo, as well as being easily enabled in others, such as SUSE, Debian, and Ubuntu. SELinux enables administrators to further harden their Linux systems and applications, making it much harder for intruders and malicious actors to abuse the system. SELinux System Administration – Third Edition provides end-to-end coverage of SELinux on Linux systems, ranging from understanding what SELinux is and how it acts to tuning SELinux controls and its integrations within Linux and application platforms, up to the definition and maintenance of custom policies.

Who this book is for This book is for Linux administrators who want to control the secure state of their systems. It's packed with the latest information on SELinux operations and administrative procedures so you'll be able to further harden your system through mandatory access control (MAC) – a security strategy that has been shaping Linux security for years. The book can also be enlightening for IT architects to understand how SELinux can be positioned to enhance the security of Linux systems and Linux-hosted services within their organization. Readers should have reasonable experience with maintaining Linux systems, covering user management, software installation and maintenance, regular Linux security controls, and network configuration.

What this book covers Chapter 1, Fundamental SELinux Concepts, provides fundamental insights into the SELinux technology and allows you to understand the differences between SELinux implementations.

xii

Preface

Chapter 2, Understanding SELinux Decisions and Logging, teaches you how to analyze SELinux events, and how to configure system logging to facilitate SELinux troubleshooting. Chapter 3, Managing User Logins, allows you to manage Linux users and associate them with the right SELinux context. Chapter 4, Using File Contexts and Process Domains, explains how SELinux labels are exposed on the system, and how to change the SELinux context of files and resources. Chapter 5, Controlling Network Communications, introduces SELinux access control protections on a network level, ranging from socket-based protection measures to packet filtering with SELinux. Chapter 6, Configuring SELinux through Infrastructure-as-Code Orchestration, shows you how to configure SELinux settings across large-scale environments using automation and orchestration tooling. Chapter 7, Configuring Application-Specific SELinux Controls, explains how SELinux is adopted by several applications to augment their security posture further. Chapter 8, SEPostgreSQL – Extending PostgreSQL with SELinux, helps you learn how to enable SEPostgreSQL in a regular PostgreSQL deployment, and how to use the SELinux controls within the database engine. Chapter 9, Secure Virtualization, uses libvirt and other virtualization technologies together with SELinux to further protect and isolate virtual guests from each other. Chapter 10, Using Xen Security Modules with FLASK, teaches you how Xen uses an SELinux-like approach to isolate its guests using Xen Security Modules, and how administrators can tweak and tune isolation further. Chapter 11, Enhancing the Security of Containerized Workloads, shows how container platforms such as Docker, podman, and Kubernetes use SELinux as a means to secure the host system from potentially untrusted containers and provide isolation between containers. Chapter 12, Tuning SELinux Policies, expands on SELinux booleans and their effect on the system and shows how to deal with different SELinux policy modules. Chapter 13, Analyzing Policy Behavior, explains how administrators and analysts can interpret the SELinux policy and use policy analysis tooling to learn what a policy will allow. Chapter 14, Dealing with New Applications, informs you how to apply SELinux on new applications that are not yet supported by the current SELinux policies.

Preface

xiii

Chapter 15, Using the Reference Policy, explains how to use the reference policy to create and adjust SELinux policies. Chapter 16, Developing Policies with SELinux CIL, introduces you to the Common Intermediate Language and how to apply it to develop custom policies.

To get the most out of this book This book focuses on the SELinux technology, which is enabled by default on many Linux distributions. While the book uses CentOS version 8 for most of its examples, any Linux distribution that has an SELinux implementation based upon the reference policy (which is the case for all major Linux distributions out there) suffices to follow this book. In Chapter 5, Controlling Network Communications, one section covers using SELinux for InfiniBand infrastructure, which requires specialized hardware if you want to follow the examples to the letter. However, most of the chapter can be followed without additional requirements. If you are using the digital version of this book, we advise you to type the code yourself or access the code via the GitHub repository (link available in the next section). Doing so will help you avoid any potential errors related to the copying and pasting of code.

Download the example code files You can download the example code files for this book from GitHub at https:// github.com/PacktPublishing/SELinux-System-AdministrationThird-Edition. In case there's an update to the code, it will be updated on the existing GitHub repository. We also have other code bundles from our rich catalog of books and videos available at https://github.com/PacktPublishing/. Check them out!

Code in Action Code in Action videos for this book can be viewed at https://bit.ly/3o4paOb.

Download the color images We also provide a PDF file that has color images of the screenshots/diagrams used in this book. You can download it here: https://static.packt-cdn.com/ downloads/9781800201477_ColorImages.pdf.

xiv

Preface

Conventions used There are a number of text conventions used throughout this book. Code in text: Indicates code words in text, database table names, folder names, filenames, file extensions, pathnames, dummy URLs, user input, and Twitter handles. Here is an example: "When activating a module, the semodule command will copy those modules into a dedicated directory."

A block of code is set as follows: chain input { type filter hook input priority 0; ct state new meta secmark set tcp dport map @secmapping_in ct state new ct secmark set meta secmark ct state established,related meta secmark set ct secmark }

When we wish to draw your attention to a particular part of a code block, the relevant lines or items are set in bold: (roleattributeset cil_gen_require system_r) (block pgpool (type domain) (roletype .system_r domain) )

Any command-line input or output is written as follows: $ cat /etc/passwd cat: /etc/passwd: Permission denied

Bold: Indicates a new term, an important word, or words that you see onscreen. For example, words in menus or dialog boxes appear in the text like this. Here is an example: "This can be accomplished through the Open Policy button, or by navigating to File | Open Policy." Tips or important notes Appear like this.

Preface

xv

Get in touch Feedback from our readers is always welcome. General feedback: If you have questions about any aspect of this book, mention the book title in the subject of your message and email us at [email protected]. Errata: Although we have taken every care to ensure the accuracy of our content, mistakes do happen. If you have found a mistake in this book, we would be grateful if you would report this to us. Please visit www.packtpub.com/support/errata, selecting your book, clicking on the Errata Submission Form link, and entering the details. Piracy: If you come across any illegal copies of our works in any form on the Internet, we would be grateful if you would provide us with the location address or website name. Please contact us at [email protected] with a link to the material. If you are interested in becoming an author: If there is a topic that you have expertise in and you are interested in either writing or contributing to a book, please visit authors.packtpub.com.

Reviews Please leave a review. Once you have read and used this book, why not leave a review on the site that you purchased it from? Potential readers can then see and use your unbiased opinion to make purchase decisions, we at Packt can understand what you think about our products, and our authors can see your feedback on their book. Thank you! For more information about Packt, please visit packt.com.

Section 1: Using SELinux

In this part, you will learn what SELinux is and how it acts on systems, as well as how to configure and manipulate SELinux essentials during system administration. This section comprises the following chapters: • Chapter 1, Fundamental SELinux Concepts • Chapter 2, Understanding SELinux Decisions and Logging • Chapter 3, Managing User Logins • Chapter 4, Using File Contexts and Process Domains • Chapter 5, Controlling Network Communications • Chapter 6, Configuring SELinux through Infrastructure-as-Code Orchestration

1

Fundamental SELinux Concepts Security-Enhanced Linux (SELinux) brings additional security measures to your Linux system to further protect its resources. As part of the Linux kernel, it is a mandatory access control system supported by major Linux distributions. In this book, we cover all aspects of SELinux, from basic fundamentals to resolving SELinux issues, configuring applications to deal with SELinux, and even writing our own policies. Before we embark on the details of SELinux, let's first cover the concepts of this technology: why SELinux uses labels to identify resources, how SELinux differs from traditional Linux access controls, how SELinux enforces security rules, and other mandatory access control systems that are supported in the Linux kernel. We will also see how the access control rules enforced by SELinux are provided through policy files. At the end of the chapter, we will cover an overview of the differences between SELinux implementations across Linux distributions. In this chapter, we're going to cover the following main topics: • Providing more security for Linux • Labeling all resources and objects • Defining and distributing policies • Distinguishing between policies

20

Fundamental SELinux Concepts

Technical requirements Check out the following video to see the Code in Action: https://bit.ly/2FFaUdm

Providing more security for Linux Seasoned Linux administrators and security engineers already know that they need to put some trust in the users and processes of their system in order for the system to remain secure. This is partly because users can attempt to exploit vulnerabilities found in the software running on the system, but a large contribution to this trust level is because the secure state of the system depends on the behavior of the users. A Linux user with access to sensitive information could easily leak that out to the public, manipulate the behavior of the applications they launch, and do many other things that affect the security of the system. The default access controls active on a regular Linux system are discretionary; it is up to the users how the access controls should behave. The Linux discretionary access control (DAC) mechanism is based on the user and/ or group information of the process and is matched against the user and/or group information of the file, directory, or other resource being manipulated. Consider the / etc/shadow file, which contains the password and account information of the local Linux accounts: $ ls -l /etc/shadow -rw-r-----. 1 root root 1019 Nov 28 20:44 /etc/shadow

Without additional access control mechanisms in place, this file is readable and writable by any process owned by the root user, regardless of the purpose of the process on the system. The shadow file is a typical example of a sensitive file that we don't want to see leaked or abused in any other fashion. Yet the moment someone has access to the file, that user can copy it elsewhere, for example to a home directory, or even mail it to another computer and attempt to attack the password hashes stored within.

Providing more security for Linux

21

Another example of how Linux DAC requires trust from its users is the configuration of a database server. Database files themselves are (hopefully) only accessible to the runtime account of the database management system (DBMS) itself, and the Linux root user. Properly secured systems will only grant trusted users access to these files (for instance, through sudo) by allowing them to change their effective user ID from their personal user to the database runtime user or even the root account, but only for a well-defined set of commands that the system administrator has configured up front. These users too, can analyze the database files and gain access to potentially confidential information in the database without going through the DBMS. Administrators often have to put significant trust in these users to provide a secure system, rather than being able to enforce this. However, regular users are not the only reason for securing a system. Lots of software daemons run as the Linux root user or have significant privileges on the system. Errors within those daemons can easily lead to information leakage or might even lead to remotely exploitable vulnerabilities. Backup software, monitoring software, change management software, scheduling software, and so on: they all often run with the highest privileged account possible on a regular Linux system. Even when the administrator does not allow privileged users, their interaction with daemons introduces a potential security risk. So, the users are still trusted to correctly interact with these applications in order for the system to function properly. Through this, the administrator leaves the security of the system to the discretion of its (many) users. Enter SELinux, which provides an additional access control layer on top of the standard Linux DAC mechanism. SELinux provides a mandatory access control (MAC) system that, unlike its DAC counterpart, gives the administrator full control over what is allowed on the system and what isn't. It accomplishes this by supporting a policy-driven approach over what processes are and aren't allowed to do and by enforcing this policy through the Linux kernel. Mandatory means that the operating system enforces the access control, defined solely by the policy rules that the system administrator (or security administrator) has enabled. Users and processes do not have permission to change the security rules, so they cannot work around the access controls; security is not left to their discretion anymore. Considering the relational database example, a mandatory access control system would no longer require the administration to trust certain users, as it has full control over what these users can and cannot do. PostgreSQL, as we will see in Chapter 8, SEPostgreSQL – Extending PostgreSQL with SELinux, can interact with the SELinux subsystem to allow the administrator full coverage over the data access involved, even inside the database.

22

Fundamental SELinux Concepts

The word mandatory here, just like the word discretionary before, was not chosen accidentally to describe the abilities of the access control system: both are known terms in the security research field. Many security publications use these terms, including the Trusted Computer System Evaluation Criteria (TSEC) (http://csrc.nist.gov/ publications/history/dod85.pdf) standard (also known as the Orange Book) published by the Department of Defense in the United States of America in 1985. This publication has led to the Common Criteria standard for computer security certification (ISO/IEC 15408), available at http://www.commoncriteriaportal.org/cc/. Next, we'll describe how the Linux kernel is responsible for the SELinux implementation.

Introducing Linux Security Modules (LSM) Consider the example of the shadow file again. A MAC system can be configured to only allow a limited number of processes to read from and write to the file. On such specifically configured systems, a user logged on as root cannot directly access the file or even move it around. They can't even change the attributes of the file: # id uid=0(root) gid=0(root) groups=0(root),1(bin),2(daemon),3(sys), 4(adm),6(disk),10(wheel),11(floppy),26(tape),27(video) context= sysadm_u:sysadm_r:sysadm_t:s0-s0:c0.c1023 # cat /etc/shadow cat: /etc/shadow: Permission denied # chmod a+r /etc/shadow chmod: changing permissions of '/etc/shadow': Permission denied

The system enforces this through rules that describe when the contents of this file can be read, or when its attributes can be changed. With SELinux, these rules are defined in the SELinux policy and are loaded when the system boots. It is the Linux kernel itself that is responsible for enforcing the rules. Mandatory access control systems such as SELinux are supported in the Linux kernel through Linux Security Modules (LSM), a Linux subsystem called before processing a user space request. Such requests are called system calls, and Linux supports over 100 of them.

Providing more security for Linux

23

Figure 1.1 – High-level overview of how LSM integrates into the Linux kernel

LSM has been available in the Linux kernel since version 2.6, released in December 2003. It is a framework that provides hooks inside the Linux kernel at various locations, including the system call entry points. When these hooks trigger, registered security implementations such as SELinux have their functions executed automatically. In SELinux, these functions check the policy and other information before returning a go/ no-go. LSM by itself does not provide any security functionality; instead, it relies on security implementations that do the heavy lifting: the framework is modular. Within the LSM framework, two types of security modules exist: exclusive and non-exclusive modules. Two exclusive modules cannot be active simultaneously: each exclusive LSM module needs exclusive control over some kernel objects (generally those related to a security context) and is not able to deal with other LSM modules that need these objects as well. Non-exclusive modules don't have this need and can be combined (also known as stacking) at will, regardless of whether an exclusive LSM module is active or not. A major use case for stacking LSM modules is to enable different security models within containers running on the system. Right now, it is not possible to implement a different security module within a Linux container, and the security within the container falls back to the security module of the host. To support this, more and more exclusive LSM implementations (like SELinux) are working to make their implementation non-exclusive, and we can expect improvements in this area within the next year.

24

Fundamental SELinux Concepts

SELinux is one implementation that uses LSM. Several other implementations exist: • AppArmor is a mandatory access control system that has a strong focus on application-level protections (called profiles), based largely on filesystem paths. This makes AppArmor easy to understand and implement for administrators, as it does not have the complexity of abstracting rules to labels (as SELinux does). In the Labeling all resources and objects section, we explain why SELinux uses labels. AppArmor is an exclusive LSM module at the time of writing, but will most likely become non-exclusive very soon. • Smack is a mandatory access control system that uses labels on processes and resources. The labels contain security identifiers interpreted by Smack to enforce access control, requiring fewer access rules in Smack (unlike SELinux, which does not perform an interpretation of labels – excluding sensitivity – and thus requires a higher number of policy rules). Smack is an exclusive LSM module. • TOMOYO Linux is a mandatory access control system, but its access control mechanism is also easy to use for system analysis. It automatically builds up policies based on application behavior, and like AppArmor, its policies primarily use paths rather than labels. TOMOYO Linux (and its fork, AKARI) is a non-exclusive LSM module. • LoadPin is an LSM module that ensures that the Linux kernel resources (such as kernel modules and firmware) are all loaded from a single non-writable filesystem. LoadPin is a non-exclusive LSM module. • Yama is an LSM module that adds additional access controls on activities that are not sufficiently fine-grained by Linux, such as by attaching them to the memory of another process (using ptrace). Yama is a non-exclusive LSM module. • SafeSetId is an LSM module that allows finer control over which users can use setuid (switching to another user) toward another user. Rather than granting the use of setuid, SafeSetId can limit for which users this is allowed. This ensures that vulnerabilities or misconfigurations in tools such as sudo are still contained. SafeSetId is a non-exclusive LSM module. • Lockdown is an LSM module that protects the Linux kernel memory. It has two modes: in integrity mode, it prevents modifying kernel objects from user space (such as direct memory access or PCI access); in confidentiality mode, it additionally prevents extracting potentially confidential information from kernel objects. Lockdown is a non-exclusive LSM module. • The capability LSM module is, by default, enabled on systems and provides support for Linux capabilities (a set of permissions granted to a user when the user is assigned a certain capability). It is a non-exclusive LSM module.

Providing more security for Linux

25

To query the list of active LSM modules on a system, read /sys/kernel/security/ lsm: $ cat /sys/kernel/security/lsm capability,selinux

Next, we'll explain how SELinux works on top of regular Linux access controls.

Extending regular DAC with SELinux SELinux does not change the Linux DAC implementation, nor can it override denials made by the Linux DAC permissions. If a regular system (without SELinux) prevents a particular access, there is nothing SELinux can do to override this decision. This is because the LSM hooks are triggered after the regular DAC permission checks execute, a conscious design decision from the LSM project. For instance, if you need to allow an additional user access to a file, you cannot add an SELinux policy to do that for you. Instead, you will need to look into other features of Linux, such as the use of POSIX access control lists. Through the setfacl and getfacl commands, the user can set additional permissions on files and directories, opening up the selected resource to additional users or groups. As an example, let's grant a user admin read-write access to a file using setfacl: $ setfacl -m u:admin:rw /srv/backup/setup.conf

Similarly, to view the current POSIX ACLs applied to the file, use this command: $ getfacl /srv/backup/setup.conf getfacl: Removing leading '/' from absolute path names # file: srv/backup/setup.conf # owner: root # group: root user::rwuser::admin:rwgroup::r-mask::rwother::r—

This shows that the file is writable not only by its owner but also by the admin user.

26

Fundamental SELinux Concepts

Restricting root privileges The regular Linux DAC allows an all-powerful user: root. Unlike most other users on the system, the logged-on root user has all the rights needed to fully manage the entire system, ranging from overriding access controls to controlling audits, changing user IDs, managing the network, and much more. This is supported through a security concept called capabilities (for an overview of Linux capabilities, check out the capabilities manual page: man capabilities). SELinux is also able to restrict access to these capabilities in a fine-grained manner. Due to this fine-grained authorization aspect of SELinux, even the root user can be confined without impacting the operations on the system. The previous example of accessing /etc/shadow is just one example of an activity that a powerful user such as root still might not be able to perform due to the SELinux access controls in place.

Reducing the impact of vulnerabilities If one benefit of SELinux needs to be stressed, then it is its ability to reduce the impact of vulnerabilities. But this vulnerability reduction is also often misunderstood. A properly written SELinux policy confines applications so that their allowed activities are reduced to a minimum set. This least-privilege model ensures that abnormal application behavior is not only detected and audited but also prevented. Many application vulnerabilities can be exploited to execute tasks that an application is not meant to do. When this happens, SELinux will prevent this. However, there are two misconceptions about SELinux's ability to thwart exploits, namely, the impact of the policy and the exploitation itself. If the policy is not written in a least-privilege model, then SELinux might consider this non-standard behavior as normal and allow the actions to continue. For policy writers, this means that their policy rules have to be very fine-grained. Sadly, that makes writing policies very time-consuming: with more than 130 classes and over 250 permissions known to SELinux, policy rules need to take all these classes and permissions into account for each interaction. As a result, policies tend to become convoluted and harder to maintain. Some policy writers make policies more permissive than is absolutely necessary, which might result in exploits becoming successful even though the action is not expected behavior from an application's point of view. Some application policies are explicitly marked as unconfined (which we discuss in Chapter 14, Dealing with New Applications), showing that they are very liberal in their allowed permissions. Fedora, CentOS, and Red Hat Enterprise Linux even start application policies as permissive and only start enforcing access controls for those applications after a few releases (and additional testing).

Providing more security for Linux

27

The second misconception is the exploit itself. If an application's vulnerability allows an unauthenticated user to use the application services as if the user were a regular, authorized user, then SELinux will not play a role in reducing the impact of the vulnerability; it will only notice the behavior of the application itself and not of the sessions internal to the application. As long as the application itself behaves as expected (such as accessing its own files and not poking around in other filesystems), SELinux will happily allow the actions to take place. It is only when the application starts behaving erratically that SELinux stops the exploit from continuing. SELinux will prevent exploits such as remote command execution (RCE) against applications that should not be executing random commands (such as database management systems or web servers, excluding CGI-like functionality), whereas session hijacking or SQL injection attacks are not controllable through SELinux policies.

Enabling SELinux support Enabling SELinux on a Linux system is not just a matter of enabling the SELinux LSM module within the Linux kernel. An SELinux implementation contains the following: • The SELinux kernel subsystem, implemented in the Linux kernel through LSM • Libraries, used by applications that need to interact with SELinux • Utilities, used by administrators to interact with SELinux • Policies, which define the access controls themselves The libraries and utilities are bundled by the SELinux user space project (https:// github.com/SELinuxProject/selinux). Next to the applications and libraries provided by the SELinux user space project, various components on a Linux system are updated with SELinux-specific code, including the init system and several core utilities. Because SELinux isn't just a switch that needs to be toggled, Linux distributions that support it usually come with SELinux predefined and loaded: Fedora, CentOS, and Red Hat Enterprise Linux (with its derivatives, such as Oracle Linux) are well-known examples. Other supporting distributions might not automatically have SELinux enabled but can easily support it through the installation of additional packages (which is the case with Debian and Ubuntu), and others have a well-documented approach to how to convert a system to SELinux (for example, Gentoo and Arch Linux).

28

Fundamental SELinux Concepts

Throughout the book, we will show examples for Gentoo and CentOS 8 (which is based on the free software of the Red Hat Enterprise Linux releases and is sponsored by Red Hat). These two distributions have different implementation details, which allow us to demonstrate the full potential of SELinux. To ensure the commands used within this book are available, some SELinux support tools might need to be installed. On Gentoo Linux, install at least the following packages: # emerge app-admin/setools sys-apps/policycoreutils

On CentOS Linux, install at least the following packages: # yum install setools-console policycoreutils-python-utils

As packages can change over time, it is sensible to look up which package provides a particular command. Important note If the mentioned packages no longer exist or do not cover all commands, please consult your distribution's documentation on which software packages to install. Most distributions allow searching for the most appropriate package as well, such as with e-file in Gentoo, or yum whatprovides on CentOS or related distributions.

With the SELinux main functionality described, let's look at how SELinux knows what is on the system, and which abstraction it uses to allow policies to be developed for a wide set of users.

Labeling all resources and objects When SELinux has to decide whether it has to allow or deny a particular action, it makes a decision based on the context of both the subject (who is initiating the action) and the object (which is the target of the action). These contexts (or parts of the context) are mentioned in the policy rules that SELinux enforces. The context of a process is what identifies the process to SELinux. SELinux has no notion of Linux process ownership and does not care how the process is called, which process ID it has, and what account the process runs as. All it wants to know is what the context of that process is, represented to users and administrators as a label. Label and context are often used interchangeably, and although there is a technical distinction (one is a representation of the other), we will not dwell on that much.

Labeling all resources and objects

29

Let's look at an example label – the context of the current user: $ id -Z sysadm_u:sysadm_r:sysadm_t:s0-s0:c0.c1023

The id command, which returns information about the current user, is shown executing with the -Z switch (a commonly agreed upon switch for displaying additional security information obtained from the LSM-based security subsystem). It shows us the context of the current user (actually the context of the id process itself when it was executing). As we can see, the context has a string representation and looks as if it has five fields (it doesn't; it has four fields – the last field just happens to contain a colon character). SELinux developers decided to use labels instead of real process and file (or other resource) metadata for its access controls. This is different from MAC systems such as AppArmor, which uses the path of the binary (and thus the process name) and the paths of the resources to handle permission checks. The following reasons inspired the decision to make SELinux a label-based mandatory access control: • Using paths might be easier to comprehend for administrators, but this doesn't allow us to keep the context information close to the resource. If a file or directory moves or remounts, or if a process has a different namespace view on the files, then the access controls might behave differently as they look at the path instead of the file. With label-based contexts, the system retains this information and keeps controlling the resource's access properly. • Contexts reveal the purpose of the process very well. The same binary application can be launched in different contexts depending on how it got started. The context value (such as the one shown in the id -Z output earlier) is exactly what the administrator needs. With it, they know what the rights are of each of the running instances, but they can also deduce from it how the process was launched and what its purpose is. • Contexts also make abstractions of the object itself. We are used to talking about processes and files, but contexts are also applicable to less tangible resources such as pipes (inter-process communication) or database objects. Path-based identification only works as long as you can write a path. As an example, consider the following policy statements: • Allow the httpd processes to bind to TCP port 80. • Allow the processes labeled with httpd_t to bind to TCP ports labeled with http_port_t.

30

Fundamental SELinux Concepts

In the first example, we cannot easily reuse this policy when the web server process isn't using the httpd binary (perhaps because it was renamed or it isn't Apache but another web server) or when we want to have HTTP access on a different port. With the labeled approach, the binary could be called apache2 or MyWebServer.py; as long as the process is labeled with httpd_t, the policy applies. The same happens with the port definition: you can label the port 8080 with http_port_t and thus allow the web servers to bind to that port as well without having to write another policy statement.

Dissecting the SELinux context To come to a context, SELinux uses at least three, and sometimes four, values. Let's look at the context of the SSH server as an example: $ ps -eZ | grep sshd system_u:system_r:sshd_t:s0-s0:c0.c1023 2629 ? 00:00:00 sshd

As we can see, the process is assigned a context that contains the following fields: • The SELinux user system_u • The SELinux role system_r • The SELinux type (also known as the domain when we are looking at a running process) sshd_t • The sensitivity level s0-s0:c0.c1023 When we work with SELinux, knowing the contexts is extremely important. In most cases, it is the third field (called the domain or type) that is most important since the majority of SELinux policy rules (over 99 percent) consist of rules related to the interaction between two types (without mentioning roles, users, or sensitivity levels). SELinux contexts are aligned with LSM security attributes and exposed to the user space in a standardized manner (compatible with multiple LSM implementations), allowing end users and applications to easily query the contexts. An easily accessible location where these attributes are presented is within the /proc pseudo filesystem. Inside each process's /proc/ location, we find a subdirectory called attr, inside of which the following files can be found: $ ls /proc/$$/attr current exec fscreate

keycreate

prev

sockcreate

Labeling all resources and objects

31

All these files, if read, display either nothing or an SELinux context. If it is empty, then that means the application has not explicitly set a context for that particular purpose, and the SELinux context will be deduced either from the policy or inherited from its parent. The meaning of the files are as follows: • The current file displays the current SELinux context of the process. • The exec file displays the SELinux context that will be assigned by the next application execution done through this application. It is usually empty. • The fscreate file displays the SELinux context that will be assigned to the next file written by the application. It is usually empty. • The keycreate file displays the SELinux context that will be assigned to the keys cached in the kernel by this application. It is usually empty. • The prev file displays the previous SELinux context for this particular process. This is usually the context of its parent application. • The sockcreate file displays the SELinux context that will be assigned to the next socket created by the application. It is usually empty. If an application has multiple subtasks, then the same information is available in each subtask directory at /proc//task//attr.

Enforcing access through types The SELinux type (the third part of an SELinux context) of a process (called the domain) is the basis of the fine-grained access controls of that process with respect to itself and other types (which can be processes, files, sockets, network interfaces, and more). In most SELinux literature, SELinux's label-based access control mechanism is fine-tuned to say that SELinux is a type enforcement mandatory access control system: when some actions are denied, the (absence of the) fine-grained access controls on the type level are most likely to blame. With type enforcement, SELinux can control an application's behavior based on how it got executed in the first place: a web server launched by a user will run with a different type than a web server executed through the init system, even though the process binary and path are the same. The web server launched from the init system is most likely trusted (and thus allowed to do whatever web servers are supposed to do), whereas a manually launched web server is less likely to be considered normal behavior and as such will have different privileges.

32

Fundamental SELinux Concepts Important note The majority of SELinux's online resources focus on types. Even though the SELinux type is just the third part of an SELinux context, it is the most important one for most administrators. Most documentation will even just talk about a type such as sshd_t rather than a full SELinux context.

Take a look at the following dbus-daemon processes: # ps -eZ | grep dbus-daemon swift_u:swift_r:swift_dbusd_t:s0-s0:c0.c512 571 ? 00:00:01 dbus-daemon swift_u:swift_r:swift_dbusd_t:s0-s0:c0.c512 649 ? 00:00:00 dbus-daemon system_u:system_r:system_dbusd_t:s0-s0:c0.c1023 2498 ? 00:00:00 dbus-daemon

In this example, one dbus-daemon process is the system D-Bus daemon running with the aptly named system_dbusd_t type, whereas two other ones are running with the swift_dbusd_t type assigned to it. Even though their binaries are the same, they both serve a different purpose on the system and as such have a different type assigned. SELinux then uses this type to govern the actions allowed by the process toward other types, including how system_dbusd_t can interact with swift_dbusd_t. SELinux types are by convention suffixed with _t, although this is not mandatory.

Granting domain access through roles SELinux roles (the second part of an SELinux context) allow SELinux to support rolebased access controls. Although type enforcement is the most used (and known) part of SELinux, role-based access control is an important method to keep a system secure, especially from malicious user attempts. SELinux roles define which types (domains) can be accessed from the current context. These types (domains) on their part define the permissions. As such, SELinux roles help define what a user (who has access to one or more roles) can and cannot do.

Labeling all resources and objects

33

By convention, SELinux roles are defined with an _r suffix. On most SELinux-enabled systems, the administrator can assign the following SELinux roles to users: • The user_r role is meant for restricted users. This role is only allowed to have processes with types specific to end-user applications. Privileged types, including those used to switch to another Linux user, are not allowed for this role. • The staff_r role is meant for non-critical operations. This role is generally restricted to the same applications as the restricted user, but it has the ability to switch roles. It is the default role for operators to have (so as to keep those users in their least privileged role as long as possible). • The sysadm_r role is meant for system administrators. This role is very privileged, enabling various system administration tasks. However, certain end-user application types might not be supported (especially if those types are used for potentially vulnerable or untrusted software) to keep the system free from infections. • The secadm_r role is meant for security administrators. This role allows changing the SELinux policy and manipulating the SELinux controls. It is generally used when a separation of duties is needed between system administrators and system policy management. • The system_r role is meant for daemons and background processes. This role is quite privileged, supporting the various daemon and system process types. However, end-user application types and other administrative types are not allowed in this role. • The unconfined_r role is meant for end users. This role allows a limited number of types, but those types are very privileged as they allow running any application launched by a user (or another unconfined process) in a more or less unconfined manner (not restricted by SELinux rules). This role, as such, is only available if the system administrator wants to protect certain processes (mostly daemons) while keeping the rest of the system operations almost untouched by SELinux.

34

Fundamental SELinux Concepts

Other roles might exist, such as guest_r and xguest_r, depending on the distribution. It is wise to consult the distribution documentation for more information about the supported roles. The seinfo command is the most common method to obtain an overview of available roles: # seinfo –-role Roles: 9 auditadm_r object_r secadm_r … user_r

With the SELinux roles identified, let's look at how we assign roles to users.

Limiting roles through users An SELinux user (the first part of an SELinux context) is not the same as a Linux (account) user. Unlike Linux user information, which can change while the user is working on the system (through tools such as sudo or su), the SELinux policy can (and generally will) enforce that the SELinux user remains the same even when the Linux user itself has changed. Because of the immutable state of the SELinux user, we can implement specific access controls to ensure that users cannot work around the set of permissions granted to them, even when they get privileged access. An example of such an access control is the user-based access control (UBAC) feature that some Linux distributions (optionally) enable, which prevents users from accessing files of different SELinux users even when those users try to use the Linux DAC controls to grant access to each other's files. The most important feature of SELinux users, however, is that SELinux user definitions restrict which roles the (Linux) user can assume. A Linux user is first assigned to an SELinux user, which does not need to be unique: multiple Linux users can be assigned to the same SELinux user. Once set, that user cannot switch to an SELinux role not associated with that SELinux user.

Labeling all resources and objects

35

The following diagram shows the role-based access control implementation of SELinux:

Figure 1.2 – Mapping Linux accounts to SELinux users

SELinux users are, by convention, defined with an _u suffix, although this is not mandatory. The SELinux users that most distributions have available are named after the role they represent, but instead of ending with _r, they end with _u. For instance, for the sysadm_r role, we have the sysadm_u SELinux user.

36

Fundamental SELinux Concepts

Controlling information flow through sensitivities The fourth part of an SELinux context, the sensitivity, is not always present (some Linux distributions, by default, do not enable sensitivity labels, but most do). This part of the label is needed for the multilevel security (MLS) support within SELinux, which is an optional setting. Sensitivity labels allow the classification of resources and the restriction of access to those resources based on a security clearance. These labels consist of two parts: a confidentiality value (prefixed with s) and a category value (prefixed with c). In many larger organizations and companies, documents are labeled internal, confidential, or strictly confidential. SELinux can assign processes certain clearance levels for these resources. With MLS, we can configure SELinux to follow the Bell-LaPadula model, a security model characterized by no read up, no write down: based on a process's clearance level, that process cannot read anything with a higher confidentiality level nor write to (or communicate otherwise with) any resource with a lower confidentiality level. SELinux does not use internal, confidential, and other labels. Instead, it uses numbers from zero (the lowest confidentiality) to whatever the system administrator has defined as the highest value (this is configurable and set when the SELinux policy is built). Categories allow us to assign resources with one or more categories, and to define access controls across categories. One of the functionalities resulting from using categories is to support multitenancy (for example, systems hosting applications for multiple customers) within a Linux system. Multitenancy is provided by assigning a set of categories to the processes and resources of one tenant, whereas the processes and resources of another tenant get a different set of categories. When a process does not have the proper categories assigned, it cannot touch the resources (or other processes) that have other categories assigned. Important note An unwritten convention in the SELinux world is that (at least) two categories are used to differentiate between tenants. By having services randomly pick two categories for a tenant out of a predefined set of categories, while ensuring each tenant has a unique combination, these services receive proper isolation. The use of two categories is not mandatory, but services such as sVirt and Docker successfully implement this methodology.

In that sense, categories are like tags, allowing us to grant access only when the tags of the process and the target resource match. As multilevel security is not often used, the benefits of only using categories are persisted in what is called multi-category security (MCS). This is a special MLS case, which only supports a single confidentiality level (s0).

Defining and distributing policies

37

Now that we know how labels are used by SELinux policies, let's look at how SELinux policies are defined and distributed.

Defining and distributing policies Enabling SELinux does not automatically start the enforcement of access. If SELinux is enabled and it cannot find a policy, it will refuse to start because the policy defines the behavior of the system (what SELinux should allow). SELinux policies are generally distributed in a compiled form (just like with software) as policy modules. These modules are then aggregated into a single policy store and loaded in memory to allow SELinux to enforce the policy rules on the system. Important note Gentoo, a source-based meta-distribution, distributes SELinux policies as (source) code, compiled and built at install time, just like it does with other software.

The following diagram shows the relationship between policy rules (policy code), policy modules, and a policy package (which is often a one-to-one mapping toward a policy store):

Figure 1.3 – Relationship between policy rules (code), policy modules, and policy store

38

Fundamental SELinux Concepts

As we can see, policies are first written, then compiled in modules, after which they are bundled and distributed. The next few sections describe each of these phases in detail.

Writing SELinux policies An SELinux policy writer can write down the policy rules in three possible languages: • In standard SELinux source format – a human-readable and well-established language for writing SELinux policies • In reference policy style, which extends the standard SELinux source format with M4 macros to facilitate the development of policies • In the SELinux common intermediate language (CIL) – a computer-readable (and with some effort, human-readable) format for SELinux policies Most SELinux supporting distributions base their policy on the reference policy (https://github.com/SELinuxProject/refpolicy/), a fully functional SELinux policy set managed as a free software project. This allows distributions to ship with a functional policy set rather than having to write one themselves. Many project contributors are distribution developers, trying to push changes of their distribution to the reference policy project itself, where the changes are peer-reviewed to ensure no rules are brought into the project that might jeopardize the security of any platform. Writing policies without the extensive set of M4 macros offered by the reference policy project is hard, which is why the reference policy has become the de facto source for policies. The SELinux CIL format is reasonably recent, and although it is very much in use already (the SELinux user space converts everything to CIL in the background), it is not that common yet for policy writers to use it directly. To show the differences between these three languages, consider the web server rule we discussed earlier, repeated here for your convenience: allow the processes labeled with httpd_t to bind to TCP ports labeled with http_port_t. In the standard SELinux source format, we write this down as follows: allow httpd_t http_port_t : tcp_socket { name_bind };

Using reference policy style, this rule is part of the following macro call: corenet_tcp_bind_http_port(httpd_t)

In the CIL language, the rule expression is like so: (allow httpd_t http_port_t (tcp_socket (name_bind)))

Defining and distributing policies

39

In most representations, we can see what the rule is about: • The subject (who is taking the action); in this case, this is the set of processes labeled with the httpd_t type. • The target resource or object (the target for the action); in this case, it is the set of TCP sockets (tcp_socket) labeled with the http_port_t type. In reference policy style, this is implied by the function name. • The action or permission; in this case, it is the action of binding to a port (name_ bind). In reference policy style, this is implied by the function name. • The result that the policy will enforce; in this case, it is that the action is allowed (allow). In reference policy style, this is implied by the function name. A policy is generally written for an application or set of applications. So, the preceding example will be part of the policy written for web servers. Policy writers will generally create three files per application or application set: • A .te file, which contains the type enforcement rules. • A .if file, which contains interface and template definitions, allowing policy writers to easily use the newly-generated policy rules to enhance other policies. You can compare this to header files in other programming languages. • A .fc file, which contains file context expressions. These are rules that assign labels to resources on the filesystem. A finished policy is then packaged into an SELinux policy module.

Distributing policies through modules Initially, SELinux used a single, monolithic policy approach: all possible access control rules were maintained in a single policy file. It quickly became clear that this is not manageable in the long term, and the idea of developing a modular policy approach was born. Within the modular approach, policy developers can write isolated policy sets for a particular application (or set of applications), roles, and so on. These policies then get built and distributed as policy modules. Platforms that need access controls for a particular application load the SELinux policy module that defines the access rules for that application.

40

Fundamental SELinux Concepts

The following diagram shows the building of policy modules. It also shows where CIL comes into play, even when the policy rules themselves are not written in CIL:

Figure 1.4 – Build process from policy rule to policy store

The binary *.pp files (which are the SELinux policy modules) are considered to be written in a high-level language (HLL). Do not assume that this means they are human-readable: these files are binary files. The consideration here is that SELinux wants to support writing SELinux policies in a number of formats, which it calls high-level languages, as long as it has a parser that can convert the files into CIL. Marking the binary module formats (which in previous SELinux versions were the binary blobs loaded in memory) as high-level allows the SELinux project to introduce the distinction between high-level languages and CIL in a backward-compatible manner. When distributing SELinux policy modules, most Linux distributions place the *.pp SELinux policy modules inside /usr/share/selinux, usually within a subdirectory named after the policy store (such as targeted). There, these modules are ready for administrators to activate them. When activating a module, the semodule command will copy those modules into a dedicated directory (/var/lib/selinux/mcs/active/modules). When all modules are aggregated in a single location, the final policy binary is compiled, resulting in /etc/selinux/targeted/policy/policy.32 (or some other number) and loaded in memory. On CentOS, the SELinux policies are provided by the selinux-policy-targeted (or -minimum or -mls) package. On Gentoo, they are provided by the various sec-policy/selinux-* packages (Gentoo uses separate packages for each module, reducing the number of SELinux policies loaded on an average system).

Distinguishing between policies

41

Bundling modules in a policy store A policy store contains a single comprehensive policy, and only a single policy can be active on a system at any point in time. Administrators can switch policy stores, although this often requires rebooting the system and might even require relabeling the entire system (relabeling is the act of resetting the contexts on all files and resources available on that system). The active policy on the system can be queried using sestatus (an SELinux status) as follows: # sestatus | grep "Loaded policy name" Loaded policy name: mcs

In this example, mcs is the currently loaded policy (store). The policy name that SELinux will use upon its next reboot is defined in the /etc/selinux/config configuration file as the SELINUXTYPE parameter. The system's init system (be it a SysV-compatible init system or systemd) is generally responsible for loading the SELinux policy, effectively activating SELinux support on the system. The init system reads the configuration, locates the policy store, and loads the policy file in memory. If the init system does not support this (in other words, it is not SELinux-aware) then the policy should be loaded through the load_ policy command. As we now have a better view of the flow used in policy development and distribution, let's see how Linux distributions can differentiate their SELinux offering.

Distinguishing between policies The most common SELinux policy store names are strict, targeted, mcs, and mls. None of the names assigned to policy stores are fixed though, so it is a matter of convention. Hence, we recommend consulting the distribution documentation to verify what the proper name of the policy should be. Still, the name often provides some information about the SELinux options enabled through the policy.

Supporting MLS One of the options that can be enabled is MLS support. The SELinux context will not have a fourth field with sensitivity information in it if this option is disabled, making the contexts of processes and files look as follows: staff_u:sysadm_r:sysadm_t

42

Fundamental SELinux Concepts

To check whether MLS is enabled, it is sufficient to see whether a process context doesn't contain such a fourth field. Another way is to check the Policy MLS Status line in the output of sestatus: # sestatus | grep MLS Policy MLS status:

enabled

Yet another method would be to look into the pseudo file, /sys/fs/selinux/mls. A value of 0 means disabled, whereas a value of 1 means enabled: # cat /sys/fs/selinux/mls 1

Policy stores that have MLS enabled are generally targeted, mcs, and mls, whereas strict generally has MLS disabled.

Dealing with unknown permissions Permissions (such as read, open, and lock) are defined both in the Linux kernel and in the policy itself. However, sometimes, newer Linux kernels support permissions that the current policy does not yet understand. Take the block_suspend permission (to be able to block system suspension) as an example. If the Linux kernel supports (and checks) this permission but the loaded SELinux policy does not understand that permission yet, then SELinux has to decide how it should deal with the permission. We can configure SELinux to perform one of the following actions: • Allow every action related to an unknown permission (allow). • Deny every action related to an unknown permission (deny). • Stop and halt the system when an unknown permission is checked (reject). We configure this through the deny_unknown value. To see the state for unknown permissions, look for the Policy deny_unknown status line in sestatus: # sestatus | grep deny_unknown Policy deny_unknown status:

allowed

Administrators can set this for themselves in the /etc/selinux/semanage.conf file through the handle-unknown variable (with allow, deny, or reject).

Distinguishing between policies

43

Supporting unconfined domains An SELinux policy can be very strict, limiting applications as close as possible to their actual behavior, but it can also be very liberal in what applications are allowed to do. One of the concepts available in many SELinux policies is the idea of unconfined domains. When enabled, it means that certain SELinux domains (process contexts) are allowed to do almost anything they want (of course, within the boundaries of the regular Linux DAC permissions, which still hold) and only a select number of domains are truly confined (restricted) in their actions. Unconfined domains are introduced to allow SELinux to be active on desktops and servers where administrators do not want to fully restrict the entire system, but only a few of the applications running on it. Generally, these implementations focus on constraining network-facing services (such as web servers and database management systems) while allowing end users and administrators to roam around unrestricted. With other MAC systems, such as AppArmor, unconfinement is inherently part of the design of the system as they only restrict actions for well-defined applications or users. However, SELinux is designed to be a full mandatory access control system and thus needs to provide access control rules even for those applications that aren't the security administrator's primary focus. By marking these applications as unconfined, almost no restrictions are imposed by SELinux. We can see whether unconfined domains are enabled on the system using seinfo, by querying the policy and asking it whether the unconfined_t SELinux type is defined. On a system where unconfined domains are supported, this type will be available: # seinfo -t unconfined_t Types: 1 unconfined_t

For a system where unconfined domains are not supported, the type will not be part of the policy: # seinfo -t unconfined_t Types: 0

Most distributions that enable unconfined domains call their policy targeted, but this convention is not always followed. Hence, it is always best to consult the policy using seinfo. CentOS enables unconfined domains, whereas with Gentoo, this is a configurable setting through the unconfined USE flag.

44

Fundamental SELinux Concepts

Limiting cross-user sharing When UBAC is enabled, certain SELinux types will be protected by additional constraints. This will ensure that one SELinux user cannot access the files (or other specific resources) of another user, even when those users are sharing their data through the regular Linux permissions. UBAC provides some additional control over information flow between resources, but it is far from perfect. Essentially, it is made to isolate SELinux users from one another. Important note A constraint in SELinux is an access control rule that uses all parts of a context to make its decision. Unlike type enforcement rules, which are purely based on the type, constraints can take the SELinux user, SELinux role, or sensitivity label into account. Constraints are generally developed once and left untouched – most policy writers will not touch constraints during their development efforts.

Many Linux distributions, including CentOS, disable UBAC. Gentoo allows users to decide whether they want UBAC through the Gentoo ubac USE flag (which is enabled by default).

Incrementing policy versions While checking the output of sestatus, we see that there is also a reference to a policy version: # sestatus | grep version Max kernel policy version:

32

This version has nothing to do with the versioning of policy rules but with the SELinux features that the currently running kernel supports. In the preceding output, 32 is the highest policy version that the running kernel supports. Every time a new feature is added to SELinux, the version number is increased. We can find the policy file itself (which contains all the SELinux rules loaded at boot time by the system) in /etc/selinux/ targeted/policy (where targeted refers to the policy store used, so if the system uses a policy store named mcs, then the path will be /etc/selinux/mcs/policy). If multiple policy files exist, use seinfo to discover which policy version file is used: # seinfo | grep Version Policy version:

31 (MLS enabled)

Distinguishing between policies

45

A list of policy feature enhancements and the Linux kernel version in which that given feature is introduced is provided next. Many of the features are only of concern to policy developers, but knowing the evolution of the features gives us a good idea about the evolution of SELinux: • Version 12 represents the "old API" for SELinux, which is now deprecated. • Version 15, introduced in Linux 2.6.0, provided the new API for SELinux. • Version 16, introduced in Linux 2.6.5, added support for conditional policy extensions. • Version 17, introduced in Linux 2.6.6, added support for IPv6. • Version 18, introduced in Linux 2.6.8, added support for fine-grained netlink socket permissions. • Version 19, introduced in Linux 2.6.12, added support for MLS. • Version 20, introduced in Linux 2.6.14, reduced the size of the access vector table. • Version 21, introduced in Linux 2.6.19, added support for MLS range transitions. • Version 22, introduced in Linux 2.6.25, added policy capabilities. • Version 23, introduced in Linux 2.6.26, added support for per-domain permissive mode. • Version 24, introduced in Linux 2.6.28, added support for explicit hierarchy (type bounds). • Version 25, introduced in Linux 2.6.39, added support for filename-based transitions. • Version 26, introduced in Linux 3.0, added support for role-transitions for non-process classes, as well as support for role attributes. • Version 27, introduced in Linux 3.5, added support for the flexible inheritance of the SELinux user and SELinux role for newly-created objects. • Version 28, introduced in Linux 3.5, added support for the flexible inheritance of the SELinux type for newly-created objects. • Version 29, introduced in Linux 3.14, added support for attributes within SELinux constraints. • Version 30, introduced in Linux 4.3, added support for extended permissions, implemented first on ioctl controls. It also introduced enhanced SELinux Xen support.

46

Fundamental SELinux Concepts

• Version 31, introduced in Linux 4.13, added support for InfiniBand access controls. • Version 32, introduced in Linux 5.5, added support for automatically deducing the intersection in sensitivity labels, called greatest lower bound, largest upper bound (glblub). By default, when an SELinux policy is built, the highest supported version as defined by the Linux kernel and libsepol (the library responsible for building the SELinux policy binary) is used. Administrators can force a version to be lower using the policy-version parameter in /etc/selinux/semanage.conf.

Different policy content Besides the policy capabilities described in the previous section, the main difference between policies (and distributions) is the policy content itself. We already covered that most distributions base their policy on the reference policy project. Although the reference policy project is considered the master for most distributions, each distribution has its own set of deviations from this main policy set. Many distributions make extensive additions to the policy without directly passing the policies to the upstream reference policy project. There are several possible reasons why this is not directly done: • The policy enhancements or additions are still immature: Fedora, CentOS, and Red Hat initially start with active, permissive policies, meaning the policies are not enforced. Instead, SELinux logs what it would have prevented and, based on those logs, the policies are then enhanced. This means that a policy is only ready after a few releases. • The policy enhancements or additions are too specific to the distribution: If a policy set is not reusable for other distributions, then some distributions will opt to keep those policies to themselves as the act of pushing changes to upstream projects takes quite some effort. • The policy enhancements or additions haven't followed the upstream rules and guidelines: The reference policy has a set of guidelines that policies need to adhere to. If a policy set does not comply with these rules, then the reference policy will not accept the contribution. • The policy enhancements or additions are not implementing the same security model as the reference policy project wants: As SELinux is a very extensive mandatory access control system, it is possible to write completely different policies. • The distribution does not have the time or resources to push changes upstream.

Summary

47

This means that SELinux policies can differ between distributions (and even releases of the same distribution). With this, we can conclude on some of the differentiation that distributions can put into their SELinux policies: they can opt to enable or disable MLS support, allow or deny unknown permissions, add distribution-provided unconfined domains, support user-based access controls, and/or deviate from the reference policy project to suit the distribution's principles.

Summary In this chapter, we saw that SELinux offers a more fine-grained access control mechanism on top of the Linux access controls. SELinux is implemented through Linux Security Modules and uses labels to identify its resources and processes based on ownership (user), role, type, and even the security sensitivity and categorization of the resource. We covered how SELinux policies are handled within an SELinux-enabled system and briefly touched upon how policy writers structure policies. Linux distributions implement SELinux policies, which can differ between distributions based on supported features, such as sensitivity labels, the default behavior for unknown permissions, support for confinement levels, or specific constraints put in place, such as UBAC. However, most of the policy rules themselves are similar and are even based on the same upstream reference policy project. Switching between SELinux enforcement modes and understanding the log events that SELinux creates when it prohibits certain access is the subject of our next chapter. In it, we will also cover how to approach the often-heard requirement of disabling SELinux, and why doing so is the wrong way forward.

Questions 1. What is the most important difference between a DAC and a MAC system? 2. How does Linux support the different MAC technologies? 3. What four fields constitute an SELinux context? 4. How does SELinux support role-based access controls? 5. Why isn't there a single SELinux policy for all Linux distributions?

2

Understanding SELinux Decisions and Logging Once we enable SELinux on the system, it starts its access control functionality, as described in the previous chapter. Once it starts, administrators need to keep a close eye on its actions, and often need to deal with unexpected behavior if one or more applications are not acting according to the SELinux policy. Through SELinux logging, we learn how SELinux enforces its policies toward the applications on the system. Administrators have to know how to switch between SELinux in full-enforcement mode (resembling a host-based intrusion prevention system) versus its permissive, logging-only mode, and use its various methods to toggle the SELinux state (enabled or disabled; permissive or enforcing). Furthermore, we should know how to disable SELinux's enforcement for a single domain rather than an entire system, and learn to interpret the SELinux log events that describe which activities SELinux has prevented. We will finish with an overview of common methods for analyzing these logging events in day-to-day operations.

50

Understanding SELinux Decisions and Logging

In this chapter, we're going to cover the following main topics: • Switching SELinux on and off • SELinux logging and auditing • Getting help with denials

Technical requirements Check out the following video to see the Code in Action: https://bit.ly/3dFaUXm

Switching SELinux on and off This is perhaps a weird section to begin with, but disabling SELinux is a commonly requested activity. Some vendors do not support their application running on a platform that has SELinux enabled, as those vendors do not have the expertise to develop SELinux policies for their own applications, or are not able to educate their own support lines to deal with SELinux. Furthermore, system administrators are generally reluctant to use security controls they do not understand or find too complex to maintain. Luckily, SELinux is becoming a de facto standard technology in several Linux distributions, which is increasing its exposure and understanding among administrators. SELinux is also capable of selectively disabling its access controls for a part of a system rather than requiring us to disable it for a complete system.

Setting the global SELinux state SELinux supports three major states that it can be in: disabled, permissive, and enforcing. These states are set in the /etc/selinux/config file, through the SELINUX variable, as illustrated in the following code snippet: $ grep ^SELINUX= /etc/selinux/config SELINUX=enforcing

Switching SELinux on and off

51

When the init system process loads the SELinux policy, the SELinux code checks the state that the administrator has configured. The states are described as follows: • If the state is disabled, then the SELinux code disables further support, booting the system further without activating SELinux. • If the state is permissive, then SELinux is active but will not enforce its policy on the system. Instead, SELinux will report any violation against the policy, but will not prevent the action itself. This is sometimes called host intrusion detection as it works in reporting-only mode. • If the state is enforcing, then SELinux is active and will enforce its policy on the system. Violations are reported and denied. This is sometimes called host intrusion prevention, as it enforces the rules while logging the actions it takes. We can use the getenforce command or the sestatus command to get information about the current state of SELinux, like so: $ sestatus | grep mode Current mode: enforcing $ getenforce Enforcing

It is also possible to query the /sys/fs/selinux/enforce pseudo-file to get similar information. If the file returns 1, then SELinux is in enforcing mode. If it returns 0, then it is in permissive mode. The following code snippet shows SELinux in enforcing mode: $ cat /sys/fs/selinux/enforce 1

When we change the /etc/selinux/config file, then we need to reboot the system for the changes to take effect. However, if we boot a system without SELinux support (disabled), re-enabling SELinux support alone will not suffice: the administrator will need to make sure that all files on the system are relabeled (the context of all files needs to be set). Without SELinux support, Linux will create and update files without updating or setting the SELinux labels on those files. When the system is later rebooted with SELinux support, SELinux will not have any knowledge of the context of a file unless the labels are reset. Relabeling the filesystem is covered in Chapter 4, Using File Contexts and Process Domains.

52

Understanding SELinux Decisions and Logging

In many situations, administrators often want to disable SELinux when it starts preventing certain tasks. This is careless to say the least, and here's why: • SELinux is a security component—part of the operating system. Disabling SELinux is like disabling a firewall because it is blocking some communication. It might help because it's a faster way of getting something to work again, but you're removing measures that were enabled to protect you. • Just as with a firewall, SELinux is configurable by rules. If an application is prevented from working correctly, we need to update the rules for that application, just as with additional firewall rules that enable a particular network flow. We will start updating SELinux policy rules from Chapter 5, Controlling Network Communications, onward. • In the worst case, when we want to allow every action an application performs unconditionally, we can still leave SELinux on and just run this application in an unrestricted SELinux domain, called a permissive domain. Distributions put significant effort in the integration of SELinux within their products, and they have awesome support channels to help you out if all things fail.

Switching to permissive or enforcing mode Most distribution-provided Linux kernels allow switching between enforcing and permissive mode through a simple administrative command. This feature is called the SELinux development mode and is supported by the CONFIG_SECURITY_SELINUX_ DEVELOP kernel configuration parameter. This kernel parameter, if set, also has the Linux kernel boot in permissive mode first, unless a specific boot option (enforcing=1) is set. Although we could consider this development mode a risk (all a malicious person would need to do is switch SELinux to permissive mode to disable its access controls), switching the mode requires strong administrative privileges (like the root user has), which most application domains don't have. The command to switch between permissive mode and enforcing mode is the setenforce command. It takes a single argument: 0 (permissive) or 1 (enforcing). The permissive and enforcing strings are allowed by the command as well. The change takes effect immediately. For instance, we can use the following command to switch to permissive mode: # setenforce 0

Switching SELinux on and off

53

The effect of setenforce is the same as writing the right integer value into the /sys/ fs/selinux/enforce pseudo-file, as illustrated in the following code snippet: # echo 0 > /sys/fs/selinux/enforce

Switching between permissive and enforcing mode can be of interest for policy developers or system administrators who are modifying the system to use SELinux properly. We can also use it to quickly verify whether an application warning or error is due to SELinux access controls or not—assuming the application is not SELinux-aware, which we will talk about in the Understanding SELinux-aware applications section. On production systems, it might be of interest to disable the ability to switch to permissive mode. Disabling this feature usually requires the Linux kernel to be rebuilt, but SELinux policy developers have also thought of a different way to disallow users from toggling the SELinux state. The privileges that users need to switch to permissive mode are conditional, and system administrators can easily toggle this to disable switching back from enforcing mode to permissive mode. The condition is implemented through an SELinux Boolean called secure_mode_policyload whose default value is off (meaning switching SELinux state is allowed). SELinux Booleans are configurable options that take on a single value (on or off, although true/false and 1/0 are valid values as well) and manipulate parts of the active SELinux policy. The value of the conditionals can be persisted (meaning they survive reboots) or be kept only during the current boot session. We can persist the value across reboots by adding -P to the setsebool command, as follows: # setsebool -P secure_mode_policyload on

SELinux Booleans are covered in more depth in Chapter 12, Tuning SELinux Policies. The use of the secure_mode_policyload SELinux Boolean allows administrators to restrict switching from enforcing mode back to permissive mode. This does not disable SELinux completely, but only toggles whether it will act upon its policies or not. Switching from a disabled state to a running state is not supported. However, the reverse is possible, but only under the following condition: if the Linux kernel is built with the SECURITY_SELINUX_DISABLE kernel configuration parameter, then services such as init can effectively disable SELinux at runtime, but only if no SELinux policy is loaded yet. This functionality, however, is not recommended to be actively used, and was only introduced for platforms where boot options are hard to use. The feature is marked as deprecated in recent kernels as such platforms are few in number.

54

Understanding SELinux Decisions and Logging

Using kernel boot parameters Using the setenforce command makes sense when we want to switch to permissive or enforcing mode at a point in time when we have interactive access to the system. But what if we need this on system boot? If the system refuses to boot properly due to SELinux access controls, we cannot edit the /etc/selinux/config file. Luckily, we can change the SELinux state through other means as well. The solution is to use kernel boot parameters. We can boot a Linux system with one or two parameters that take precedence over the /etc/selinux/config setting, as follows: • selinux=0: This informs the system to disable SELinux completely, and has the same effect as setting SELINUX=disabled in the configuration file. When set, the other parameter (enforcing) is not consulted. Please remember that booting a system with SELinux disabled means that to enable it again, we need to relabel all files and resources on the filesystem. The selinux= parameter is supported through the CONFIG_SECURITY_SELINUX_BOOTPARAM kernel configuration. • enforcing=0: This informs the system to run SELinux in permissive mode, and has the same effect as setting SELINUX=permissive in the configuration file. • enforcing=1: This informs the system to run SELinux in enforcing mode, and has the same effect as setting SELINUX=enforcing in the configuration file. Consider a Linux system that uses GRUB2 as its boot loader, and we want to add enforcing=0 to the boot entry. To accomplish this, we execute the following steps: 1. Reboot the system until the GRUB2 boot screen comes up. 2. Navigate with the arrow keys to the boot entry for which the SELinux state must be altered. This is usually the default boot entry and should be already selected. 3. Press the E key to edit the boot entry line. Do this before the GRUB2 timer reaches zero; otherwise, the system will continue to boot. 4. Use the arrow keys to go to the end of the line that starts with options. If no such line exists, go to the end of the line that starts with linux, linux16, or linuxefi. 5. Add enforcing=0 to the end of this line. 6. Press Ctrl + X or F10 to boot the entry. Other boot loaders have similar approaches to changing the boot line without persisting it for every reboot. Consult your distribution documentation for more details.

Switching SELinux on and off

55

Alongside the SELinux-specific parameters, there are a few Linux Security Module (LSM)-related boot parameters that can be useful to know, especially when you are combining multiple LSM modules on the same system. These are detailed as follows: • The lsm.debug boot parameter enables LSM initialization debugging output, showing which LSM modules it effectively enables or ignores, and which LSM modules are considered as exclusive. • The lsm=lsm1,…,lsmN option chooses the order of LSM initialization. For instance, to initialize SELinux before lockdown, use lsm=selinux,lockdown. • The security= boot parameter enables selection of the active, major/exclusive LSM module. This parameter, however, is deprecated, favoring the lsm= parameter. When using SELinux in production, it might be wise to properly protect the boot menu— for instance, by password-protecting the menu and regularly verifying the integrity of the boot menu files.

Disabling SELinux protections for a single service Since policy version 23 (which came with Linux 2.6.26), SELinux also supports a more granular approach to switching between permissive and enforcing mode: the use of permissive domains. As mentioned before, a domain is a term that SELinux uses for types (labels) assigned to processes. With permissive domains, we can mark one or more domains as permissive (and, as such, not enforced by SELinux rules), even though the rest of the system is still running in enforcing mode. To make a domain permissive, we use the semanage command, as follows: # semanage permissive -a minidlna_t

With the same semanage command, we can list the currently defined permissive domains, like this: # semanage permissive -l Builtin Permissive Types Customized Permissive Types minidlna_t

56

Understanding SELinux Decisions and Logging

In the previous example, you will notice that there is also room for built-in permissive types. These are domains that have been marked as permissive by the policy developers of the Linux distribution itself. Some distributions opt to introduce new application policies in permissive mode first, allowing users to test out the policies before enforcing them. When that is the case, you can find these permissive domains under Builtin Permissive Types. Another method for listing the custom permissive types (those not marked as permissive through the distribution) is to use the semodule command. In the previous chapter, we briefly touched on this command when talking about SELinux policy modules. We can use it to list the SELinux policy modules that have permissive_ in their name because the semanage permissive command generates a small SELinux policy module to mark the domain as permissive, as illustrated in the following code snippet: # semodule -l | grep permissive_ permissive_minidlna_t

To remove the permissive mode from the domain, pass the -d argument to the semanage command. This is only possible for domains that the system administrator marked as permissive, though—distribution-provided permissive domains cannot be switched to enforcing mode through this approach. This is illustrated in the following code snippet: # semanage permissive -d minidlna_t

When a domain is marked as permissive, the application should behave as if SELinux is not enabled on the system (SELinux will not be enforcing anything that particular application/domain does), making it easier for us to discover whether SELinux is really causing a permission issue. Note, though, that other domains (including those that interact with a permissive domain) are themselves still governed and enforced through the SELinux access controls.

Understanding SELinux-aware applications Most applications themselves do not have knowledge that they are running on an SELinux-enabled system. Without this knowledge, permissive mode truly means that the application behaves as if SELinux were not enabled to begin with. However, some applications actively rely on the SELinux policy to make access control decisions, or interact with SELinux for further information gathering. We call these applications SELinux-aware because they change their behavior based on the SELinux-related information available.

SELinux logging and auditing

57

Sadly, many of these SELinux-aware applications do not properly validate whether they are running in permissive mode or not. As a result, running these applications in a permissive domain (or the entire system in permissive mode) will generally not result in the application running as if SELinux were not active. Examples of such applications are the Secure Shell (SSH) daemon, the system login service, the init system, and some cron daemons, as well as several core Linux utilities (such as ls and id). They might show permission failures or different behavior based on the SELinux policy, even if SELinux is not in enforcing mode. We can find out whether an application is SELinux-aware by checking whether the application is dynamically linked with the libselinux library. Such checks are possible with readelf, ldd, or objdump, as follows: $ readelf -d /bin/ls | grep selinux 0x0000000000000001 (NEEDED) Shared library: [libselinux. so.1] $ ldd /bin/ls | grep selinux libselinux.so.1 => /lib64/libselinux.so.1 (0x00005d415f3f03f0) $ objdump -x /bin/ls | grep selinux NEEDED libselinux.so.1

Knowing whether an application is SELinux-aware or not can help in troubleshooting failures, as the application's behavior might still be different between a disabled SELinux state and a permissive SELinux state. Up until now, we've focused on enabling or disabling SELinux, and thus on a granular or coarse-grained matter. Once it is enabled though, its interaction with the administrator will be through policy enforcement and logging. So, let's look at how SELinux handles logging.

SELinux logging and auditing SELinux developers understand that a security-oriented subsystem such as SELinux can only succeed if it is capable of enhanced logging and—even—debugging. Every action that SELinux takes, as part of the LSM hooks that it implements, should be auditable. Denials (actions that SELinux prevents) should always be logged so that administrators can take due action. SELinux tuning and changes, such as loading new policies or altering SELinux Booleans, should always result in an audit event.

58

Understanding SELinux Decisions and Logging

Following audit events By default, SELinux will send its messages to the Linux audit subsystem (assuming the Linux kernel is configured with the audit subsystem enabled through the CONFIG_AUDIT kernel configuration). There, the messages are picked up by the Linux audit daemon (auditd) and logged in the /var/log/audit/audit.log file. Distributions and administrators can define additional handling rules by configuring the audit dispatcher process (audisp), which picks up audit events and dispatches them to one or more separate processes. The SELinux troubleshooting daemon (setroubleshootd), an optional service to provide help with troubleshooting SELinux events, uses this to get access to audit events. The audit event flow is shown in this diagram:

Figure 2.1 – Flow of audit events generated by SELinux

SELinux logging and auditing

59

With SELinux enabled, (almost) every permission check that results in a denial is logged. When Linux auditing is enabled, these denials are logged by the audit daemon in the audit.log file by default. If the audit daemon is unavailable, the events are stored in the Linux kernel message buffer, which we can consult using the dmesg command. The events in the kernel message buffer are also often captured through the system logger. If the SELinux troubleshooting daemon is installed, then the audit daemon will, alongside its logging, also dispatch the events through the audit dispatch system toward the sedispatch command. This command will further handle the event and send it through D-Bus (a system bus implementation popular on Linux systems) to the SELinux troubleshooting daemon. This daemon will analyze the event and might suggest one or more fixes to the administrator. We will cover the SELinux troubleshooting daemon in the Getting help with denials section. Whenever SELinux verifies a particular access, it does not always go over the entire policy. Instead, it has an access vector cache (AVC), in which it stores the results of previous access attempts. This cache ensures that SELinux can quickly react to activities without having a huge impact on performance. We notice the abbreviation of this cache as the message type for most SELinux events, as shown at the beginning of the following example: type=AVC msg=audit(03/22/2020 12:15:38.557:2331): avc: denied { read } for pid=12569 comm="dmesg" name="xterm-256color" dev="sdb2" ino=131523 scontext=sysadm_u:sysadm_r:dmesg_t:s0s0:c0.c1023 tcontext=system_u:object_r:etc_t:s0 tclass=file permissive=0

When the Linux kernel checks a permission request, this request is represented as an access vector, and the cache is then consulted to quickly find the appropriate response. If the cache has the right access vector, then the decision is taken from the cache; otherwise, the SELinux subsystem consults the policy itself and updates the cache. Of course, SELinux invalidates the cache when a new policy is loaded or the policy is dynamically adjusted. This ensures that all permission checks are in line with the active policy. This inner working of SELinux is less relevant to most administrators, but at least now we know where the term AVC comes from.

60

Understanding SELinux Decisions and Logging

Tuning the AVC The AVC can be slightly tuned, by setting the size of the cache or its related tables. We can configure the cache size itself through the /sys/fs/selinux/avc/cache_ threshold pseudo-file (available if the CONFIG_SECURITY_SELINUX_AVC_STATS kernel configuration is set). For instance, to increase the cache size to 768 entries (the default is 512), the following command would be used: # echo 768 > /sys/fs/selinux/avc/cache_threshold

To confirm the cache threshold, read the file, as follows: # cat /sys/fs/selinux/avc/cache_threshold 768

The AVC hash statistics are available through the hash_stats pseudo-file, as illustrated in the following code snippet: $ cat /sys/fs/selinux/avc/hash_stats entries: 506 buckets used: 233/512 longest chain: 5

If you suspect that lower system performance is due to SELinux, then we advise you to look at the longest chain output in hash_stats. If it is longer than 10, then some performance impact can be expected, and updating the cache size might help. The avcstat command shows the evolution of the cache over time (the first number is the total since boot). When the number of cache misses is high or volatile, or the number of reclaims (obsoleting oldest cache entries and reusing them for new ones) is volatile, then the cache size might need to be increased. The command is illustrated in the following code snippet: $ avcstat 5 lookups hits misses allocs reclaims frees 58396334 58382324 14010 14010 10736 13511 591 591 0 0 0 0 1657 1653 4 4 0 0

SELinux logging and auditing

61

Recent kernels also allow the number of buckets used through a kernel configuration parameter to be set (CONFIG_SECURITY_SELINUX_SIDTAB_HASH_BITS), and its cache statistics can be viewed through the /sys/fs/selinux/ss/sidtab_hash_ stats pseudo-file, as illustrated in the following code snippet: $ cat /sys/fs/selinux/ss/sidtab_hash_stats entries: 285 buckets use: 55/512 longest chain: 3

Another performance parameter is the size of the internal session ID (SID) (the internal identifier used to represent contexts) to the string cache. Sadly, we can only configure this parameter at kernel build time, using the CONFIG_SECURITY_SELINUX_SID2STR_ CACHE_SIZE setting.

Uncovering more logging There is an important SELinux policy directive that provides control over what is (not) audited, and that is dontaudit. A dontaudit rule in the SELinux policy tells SELinux that an access denial should not be logged. This is the only example where SELinux won't log a denial—the SELinux policy writer has explicitly disabled the auditing of events. This is usually done to remove clutter from the logs and hide cosmetic denials that have no influence on the security of the system. The seinfo utility can tell us how many of these rules, as well as its sibling rule auditallow (log events, even though they are allowed by the policy), are currently active, as illustrated in the following code snippet: $ seinfo | grep -i audit Auditallow: 1 Dontaudit: Auditallowxperm: 0 Dontauditxperm:

5559 0

Luckily, we can disable these dontaudit rules at will. Through the following semodule command, these rules are removed from the active policy: # semodule –-disable_dontaudit –-build

The arguments can also be abbreviated to -D and -B, respectively. To re-enable the dontaudit rules, just rebuild the policy like so: # semodule -B

62

Understanding SELinux Decisions and Logging

Disabling the dontaudit rules can sometimes help in troubleshooting failures that do not result in any useful audit event. Generally speaking, though, audit events that policy writers mark as cosmetic are not the cause of a failure.

Configuring Linux auditing SELinux will try to use the audit subsystem when available and will fall back to regular system logging when it isn't. This can either be because the Linux kernel audit subsystem is not configured or because the Linux audit daemon itself is not running. For a Linux audit, we usually do not need to configure anything as SELinux AVC denials are logged by default. You will find the denials in the audit log file (/var/log/audit/ audit.log), usually together with the system call and other event messages related to the same action, as illustrated in the following code snippet: type=PROCTITLE msg=audit(...) : proctitle=ping 8.8.8.8 type=SYSCALL msg=audit(...) : arch=x86_64 syscall=socket success=no exit=EACCES(Permission denied) a0=inet a1=SOCK_ DGRAM a2=icmp a3=0x7fffac013050 items=0 ppid=2685 pid=17292 auid=admin uid=root gid=root euid=root suid=root fsuid=root egid=root sgid=root fsgid=root tty=tty1 ses=1 comm=ping exe=/bin/ping subj=sysadm_u:sysadm_r:ping_t:s0-s0:c0.c1023 key=(null) type=AVC msg=audit(...) : avc: denied { create } for pid=17292 comm=ping scontext=sysadm_u:sysadm_r:ping_t:s0s0:c0.c1023 tcontext=sysadm_u:sysadm_r:ping_t:s0-s0:c0.c1023 tclass=icmp_socket permissive=0

To configure the target log file for the audit system, use the log_file parameter in / etc/audit/auditd.conf. To enable remote audit logging (to centralize audit events from multiple hosts on a single system), you have the option of either enabling syslog forwarding or enabling the audisp-remote plugin. With syslog forwarding, the audit dispatch daemon is configured to send audit events to the local system logger as well. It is then up to the administrator to configure the local system logger to pass on events toward a remote system.

SELinux logging and auditing

63

Informational note The use of syslog forwarding has the advantage that no additional software deployments and daemons are needed on the servers to centralize their log events. The setup is hence also reusable in case hardened appliances are introduced to the environment. Of course, other log management solutions exist that can watch for log events and send those to the central server. These, however, require more configuration, and introduce an additional software agent to install.

Edit the /etc/audit/plugins.d/syslog.conf file and set active to yes, as follows: # vi /etc/audit/plugins.d/syslog.conf active = yes direction = out path = /sbin/audisp-syslog type = always args = LOG_INFO format = string

Using the system logger to centralize audit events might not be the best option though, as system loggers generally use unencrypted—and often not even guaranteed—data delivery. With the audisp-remote plugin, we can even use an encrypted channel to send the audit events, and provide guaranteed delivery to a remote auditd server. First, configure the audit daemon on the target (log) server to accept audit logs from remote hosts by enabling the audit daemon to listen on port 60. We also change the event formatting to an enriched value and add hostnames to the events so that we can distinguish events from multiple hosts, as follows: # vi /etc/audit/auditd.conf tcp_listen_port = 60 log_format = ENRICHED name_format = HOSTNAME

Next, on the source systems, configure auditd.conf as shown in the previous code snippet, but without the port setting. Then, configure the audisp-remote plugin to connect to the target server's audit daemon, as follows: # vi /etc/audit/audisp-remote.conf remote_server = port = 60

64

Understanding SELinux Decisions and Logging

Finally, enable the audisp-remote plugin, as follows: # vi /etc/audit/plugins.d/au-remote.conf active = yes

Don't forget to restart the audit daemon so that the changes take effect. We can only recommend you always use the Linux audit subsystem. Not only does it integrate nicely with troubleshooting utilities; it also allows administrators to use the audit tools to query the audit logs or even generate reports, such as with aureport, as illustrated in the following code snippet: # aureport --avc --start recent AVC Report =============================================================== # date time comm subj syscall class permission obj result event =============================================================== ... 7. 03/21/2020 19:40:55 sudo sysadm_u:sysadm_r:sysadm_sudo_t:s0s0:c0.c1023 257 dir search sysadm_u:sysadm_r:sysadm_t:s0-s0:c0. c1023 denied 1067 ... 10. 03/21/2020 19:48:19 dmesg sysadm_u:sysadm_r:dmesg_t:s0s0:c0.c1023 21 file read system_u:object_r:etc_t:s0 denied 1080

The Linux audit system is an important aide for Linux administrators, and not just for SELinux troubleshooting. But next to the Linux audit system, events can also be directed toward the local system logger, as explained next.

Configuring the local system logger When auditing is not enabled, or the Linux audit daemon is not running, then the system logger is responsible for capturing SELinux events. The system logger will log these events through the kernel logging facility (kern.*). Most system loggers will save these kernel log events in a general log file, such as /var/log/messages. We can configure the system logger to direct SELinux AVC messages into its own log file, such as /var/log/avc.log. For instance, for the rsyslog system logger, we can add in a configuration entry under /etc/rsyslog.d named 99-selinux.conf, with the following content: # vi /etc/rsyslog.d/99-selinux.conf :msg, contains, "avc: " -/var/log/avc.log

SELinux logging and auditing

65

After restarting the system logger, the AVC-related messages will show up in the /var/ log/avc.log file. When the local system logger handles SELinux logging, an easy method to quickly obtain the latest AVC denials (or other messages) is through the dmesg command, as illustrated in the following code snippet: # dmesg | grep avc | tail

Be aware, though, that unlike the audit logs, many systems allow the dmesg content to be read by regular users. This might result in some information leakage to untrusted users. For this reason, some SELinux policies do not allow regular users to access the kernel ring buffer (and, as such, use dmesg) unless the user_dmesg SELinux Boolean is set to on, as illustrated in the following code snippet: # setsebool user_dmesg on

The user_dmesg SELinux Boolean is not available on CentOS, though. There, only the standard unconfined user type and the administrative user type have access to the kernel ring buffer. To prevent other users from reading this information, you need to map these users to non-administrative SELinux users, such as user_u or (x)guest_u, which is something described further on in this book.

Reading SELinux denials The one thing every one of us will have to do several times with SELinux systems is to read and interpret SELinux denial information. When SELinux prohibits access and there is no dontaudit rule in place to hide it, SELinux will log it. If nothing is logged, it was probably not SELinux that was responsible for the failure. Remember: SELinux comes after Linux discretionary access control (DAC) checks, so if a regular permission doesn't allow a certain activity then SELinux is never consulted. SELinux denial messages are logged the moment SELinux prevents some access from occurring. When SELinux is in enforcing mode, the application usually returns a Permission denied error, although sometimes it might be a bit more obscure. An example of this can be seen in the following code snippet: $ ls /proc/1 ls: cannot access '/proc/1': Permission denied # ls -ldZ /proc/1 dr-xr-xr-x. 9 root system_u:system_r:init_t:s0 0 Mar 21 10:54 / proc/1

66

Understanding SELinux Decisions and Logging

So, what does a denial message look like? The following command output shows a denial from the audit subsystem, which we can query through the ausearch command: # ausearch -m avc -ts recent -i type=AVC msg=audit(03/22/2020 12:15:38.557:2331): avc: denied { read } for pid=12569 comm="dmesg" name="xterm-256color" dev="sdb2" ino=131523 scontext=sysadm_u:sysadm_r:dmesg_t:s0s0:c0.c1023 tcontext=system_u:object_r:etc_t:s0 tclass=file permissive=0

Let's break up this denial into its individual components. The following list gives more information about each part of the preceding denials. As an administrator, knowing how to read denials is extremely important, so take enough time for this: • SELinux action: The action that SELinux took or would take if run in enforcing mode. This is usually denied, although some actions are explicitly marked to be audited as well and would result in granted. Example: denied • Permissions: The checked permissions (action initiated by the process). This is usually a single permission, although it can sometimes be a set of permissions. Example: { read } • Process ID (PID): The ID of the process that was performing the action. Example: pid=12569 • Process name: The process name (command). It doesn't display any arguments to the command, though. Example: comm="dmesg" • Target name: The name of the target (resource) that the process is performing an action on. If the target is a file, then the name is usually the filename or directory. Example: name="xterm-256color" • Target device: The device on which the target resource resides. Together with the next field (inode number) this allows us to uniquely identify the resource on a system. Example: dev="sdb2" • Target file inode number: The inode number of the target file or directory. Together with the device, this allows us to find the file on the filesystem. Example: ino=131523 • Source context: The context in which the process resides (the domain of the process). Example: scontext=sysadm_u:sysadm_r:dmesg_t:s0-s0:c0. c1023 • Target context: The context of the target resources. Example: tcontext=system_u:object_r:etc_t:s0

SELinux logging and auditing

67

• Object class: The class of the target object—for instance, a directory, file, socket, node, pipe, file descriptor, filesystem, or capability. Example: tclass=file • Permissive mode: The mode the domain was in when the action was executed. If set to 0, then SELinux was in enforcing mode; otherwise, it was permissive (either for the system or for the given domain). Example: permissive=0 We can interpret the previous denial like so: SELinux has denied the dmesg command to read a file named "xterm-256color". The file has inode number 131523 on device / dev/sdb2 and is labeled as etc_t. The dmesg command has PID 12569 and is labeled as dmesg_t. The dmesg_t domain was not in permissive mode. Depending on the action and the target class, SELinux uses different fields to give all the information we need to troubleshoot a problem. Consider the following denial: type=AVC msg=audit(03/22/20 18:12:52.177:2326): avc: denied { name_bind } for pid=15983 comm="nginx" src=89 scontext=system_u:system_r:httpd_t:s0 tcontext=system_u:object_r:reserved_port_t:s0 tclass=tcp_socket permissive=0

The preceding denial came up because the nginx web server was configured to listen on a non-default port (89 instead of the default 80). Identifying the problem is a matter of understanding how the operations work and properly reading the denials. The denial logs give us enough to get us started, giving a clear idea of what was denied. Administrators might want to update the SELinux policy to allow a specific action (by adding an allow rule to the SELinux policy, as described further on in this book). This is, however, not always the right approach because other options exist and are usually better, such as these: • Providing the right label on the target resource (usually the case when the target is a non-default port, non-default location, and so on) • Switching Booleans (flags that manipulate the SELinux policy) to allow additional privileges • Providing the right label on the source process (often the case when the acting application is not installed by the distribution package manager) • Using the application as intended instead of through other means (as SELinux only allows expected behavior), such as starting a daemon through a service (init script or systemd unit) instead of through a command-line operation

68

Understanding SELinux Decisions and Logging

If the preceding nginx example were a wanted configuration (using a non-default port), then we should label this port as a HyperText Transfer Protocol (HTTP) port and not allow the httpd_t domain to bind on (many) other ports.

Other SELinux-related event types Although most SELinux log events are AVC-related, they aren't the sole event types an administrator will have to deal with. Most audit events will show SELinux information as part of the event, even though SELinux has little to do with the event itself, but a few audit event types are directly concerned with SELinux. Tip A full list of all possible audit events is available in the linux/audit.h header file, located in /usr/include.

USER_AVC A USER_AVC event resembles regular AVC audit events, but now the source is a user space object manager. These are applications that use SELinux policy rules, but they enforce these rules themselves rather than through the kernel. The following example is such an event, generated by D-Bus: type=USER_AVC msg=audit(03/22/2020 11:25:56.123:154) : pid=540 uid=dbus auid=unset ses=unset subj=system_u:system_r:system_dbusd_t:s0-s0:c0.c1023 msg='avc: denied { acquire_svc } for service=com.redhat. tuned spid=1460 scontext=system_u:system_r:tuned_t:s0 tcontext=system_u:system_r:tunned_t:s0 tclass=dbus permissive=0 exe=/usr/bin/dbus-daemon sauid=dbus hostname=? addr=? terminal=?'

The event has two parts. Everything up to the msg= string is information about the user space object manager that generated the event, and is the first part of the event. The true event itself (which is the second part) is stored within the msg= part and includes similar fields, as we already know from regular AVCs.

SELinux logging and auditing

69

SELINUX_ERR An SELINUX_ERR event comes up when SELinux detects a general policy violation rather than an access control violation. It cannot be resolved by SELinux policy writers by just allowing the operation. These events usually point to a misuse of applications and services that the policy is not tailored to accomplish, and an example is shown in the following code snippet: type=PATH msg=audit(03/22/2020 12:25:53.104:2364) : item=0 name=/usr/sbin/rpc.nfsd inode=3019958 dev=08:12 mode=file,755 ouid=root ogid=root rdev=00:00 obj=system_u:object_r:nfsd_ exec_t:s0 nametype=NORMAL cap_fp=none cap_fi=none cap_fe=0 cap_ fver=0 cap_frootid=0 type=SELINUX_ERR msg=audit(03/22/2020 12:25:53.104:2364) : op=security_compute_sid invalid_context=sysadm_u:sysadm_r:nfsd_ t:s0-s0:c0.c1023 scontext=sysadm_u:sysadm_r:sysadm_t:s0-s0:c0. c1023 tcontext=system_u:object_r:nfsd_exec_t:s0 tclass=process

In the preceding example, a user (running in the sysadm_t domain) was executing rpc.nfsd (with nfsd_exec_t as the label), and the policy wanted to transition to the nfsd_t domain. However, that resulted in a full context of sysadm_u:sysadm_r:nfsd_t:s0-s0:c0.c1023, which is not a valid context. The sysadm_r SELinux role does not support the nfsd_t domain.

MAC_POLICY_LOAD A MAC_POLICY_LOAD event occurs whenever the system loads a new SELinux policy in memory. This occurs when the administrator loads a new or updated SELinux policy module, rebuilds the policy with the dontaudit rules disabled, or toggles an SELinux Boolean that the administrator wants to persist across reboots. Such an event is illustrated in the following code snippet: type=MAC_POLICY_LOAD msg=audit(03/22/2020 12:28:17.077:2368) : auid=admin ses=1 lsm=selinux res=yes

When a MAC_POLICY_LOAD event occurs, you might notice a subsequent USER_MAC_ POLICY_LOAD event. This occurs when a user space object manager detects an update on the SELinux policy and takes action. Note that not all user space object managers will send out this event: some object managers will query the live policy and, as such, do not need to act when a new policy loads.

70

Understanding SELinux Decisions and Logging

MAC_CONFIG_CHANGE When an SELinux Boolean changes but doesn't persist, then a MAC_CONFIG_CHANGE event will be dispatched. This tells the administrator that the active policy has been instructed to change its behavior slightly, but within the bounds of the existing loaded policy. Such an event is illustrated in the following code snippet: type=MAC_CONFIG_CHANGE msg=audit(03/22/2020 12:29:49.564:2370) : bool=virt_use_nfs val=0 old_val=1 auid=admin ses=1

In the preceding example, the virt_use_nfs SELinux Boolean was changed from the value 1 (on) to 0 (off).

MAC_STATUS A MAC_STATUS event shows up when the SELinux enforcement state has been changed. For instance, when an administrator uses setenforce 0 to put SELinux in permissive mode, then the following event occurs: type=MAC_STATUS msg=audit(03/22/2020 12:30:45.200:2372) : enforcing=0 old_enforcing=1 auid=admin ses=1 enabled=1 oldenabled=1 lsm=selinux res=yes

MAC_STATUS is also used to inform administrators when the SELinux state itself (enabled or disabled) is altered.

NetLabel events NetLabel is a Linux kernel project to support labeled network packets, allowing security contexts such as SELinux contexts to be passed on between hosts. One of the protocols that the NetLabel implementation supports in Linux is Common IP Security Option (CIPSO) labeling, which we will cover in Chapter 5, Controlling Network Communications. The following audit events are related to the NetLabel capability: • MAC_UNLBL_STCADD and MAC_UNLBL_STCDEL events are triggered when a static label is added or removed. Static labeling means that if a packet is received or sent and it does not have a label, then this "default" static label is assigned. • MAC_MAP_ADD and MAC_MAP_DEL events are triggered when a mapping between a labeling protocol (such as CIPSO) and its parameters against an LSM (SELinux) domain is added or removed from the configuration. • MAC_CIPSOV4_ADD and MAC_CIPSOV4_DEL events are triggered when a CIPSO (IPv4) configuration is added or removed.

SELinux logging and auditing

71

Labeled IPsec events Another labeled network protocol that Linux supports is labeled IPsec, where IPsec is short for Information Protocol Security. Through this, the SELinux context of the source process (which is communicating over the IPsec tunnel toward a target resource) is known by the IPsec daemons at both ends of the tunnel. Furthermore, SELinux will contain rules about which domains can communicate over an IPsec tunnel and which domains can communicate with each other network-wise. The following audit events are related to IPsec: • MAC_IPSEC_ADDSA and MAC_IPSEC_DELSA events are used when a security association is added or removed (new IPsec tunnels are defined or deleted). • MAC_IPSEC_ADDSPD and MAC_IPSEC_DELSPD events are used when a security policy definition is added or removed. Security policies generally describe whether network packets need to be handled by IPsec and, if so, through which security association. • A MAC_IPSEC_EVENT event is a generic event for IPsec audit messages. SELinux support for labeled IPsec is described further on in this book.

Using ausearch The ausearch command, part of the Linux audit framework, is a frequently used command for querying audit events stored on the system. We already briefly covered it when taking a first look at an AVC denial, but only briefly mentioning it won't do it justice. With ausearch, we can search for events that originated during or after a selected time period. We used the -ts recent (time start) option in the past, which displays events that occurred during the past 10 minutes. The argument can also be a timestamp. Other supported shorthand values are listed as follows: • today, meaning starting at 1 second past midnight on the current day • yesterday, meaning starting at 1 second past midnight the previous day • this-week, this-month, or this-year, meaning starting at 1 second past midnight on the first day of the current week, current month, or current year • checkpoint, which uses the timestamp mentioned in a checkpoint file created in a previous run

72

Understanding SELinux Decisions and Logging

• boot, which implies only events since the system booted should be shown • week-ago, meaning starting at 1 second after midnight exactly 7 days ago The use of checkpoint is particularly useful when troubleshooting SELinux issues as it allows us to show denials (and other SELinux events) since the last invocation of the ausearch command. This is illustrated in the following code snippet: # ausearch --checkpoint /root/ausearch-checkpoint.txt -ts checkpoint

This allows administrators to perform minor tweaks and reproduce the problem and only see the events since then, instead of going through all events over and over again. By default, the ausearch command displays all events stored in the audit log. On busy systems, this can be very verbose and may result in unwanted events being displayed as well. Luckily, users can limit the type of events queried through the ausearch command. For SELinux troubleshooting, using avc,user_avc,selinux_err limits the events nicely to those needed for the job, as illustrated in the following code snippet: # ausearch -m avc,user_avc,selinux_err -ts recent

If the numeric display of fields such as user IDs and timestamps is too confusing, then it is possible for ausearch to look up and translate user IDs to usernames and timestamps to formatted time fields. Add the -i option to ausearch to have it interpret these fields and display the interpreted values instead. In this section, we've seen how SELinux notifies the system about its actions through log events, and where these log events are stored. In the next section, we'll look at how to act upon these events.

Getting help with denials On some distributions, additional support tools are available that help us identify the cause of a denial. These tools have some knowledge of common mistakes (for instance, setting the right context on files to allow the web server to read them). Other distributions require us to use our experience to make proper decisions, supporting us through the distribution mailing lists, bug tracking sites, and other cooperation locations—for example, Internet Relay Chat (IRC).

Getting help with denials

73

Troubleshooting with setroubleshoot In CentOS (and other Red Hat Enterprise Linux (RHEL)-related distributions such as Fedora), additional tools are present that help us troubleshoot denials. The tools work together to catch a denial, look for a plausible solution, and inform the administrator about the denial and its suggested resolutions. When used on a graphical workstation, denials can even result in popups that ask the administrator to review them immediately. Install the setroubleshoot package to get this support. On servers without a graphical environment, administrators can see the information in the system logs or can even configure the system to send out SELinux denial messages via email. Install the setroubleshoot-server package to get this support. Under the hood, it is the audit daemon that triggers its audit event dispatcher application (audispd). This application supports plugins, something the SELinux folks gratefully implemented. They built an application called sedispatch that will act as a plugin for audispd. The sedispatch application checks whether the audit event is an SELinux denial and, if so, forwards the events to D-Bus. D-Bus then forwards the events to the setroubleshootd application (or launches the application if it isn't running yet), which analyzes the denial and prepares feedback for the administrator. When running on a workstation, seapplet is triggered to show a popup on the administrator workstation. The administrator can then select Show to view more details. Administrators don't need a graphical user interface to be informed about SELinux issues, though. You can find analyzed feedback on the filesystem, and in the system logs you can read how to easily reach this information, as illustrated in the following code snippet: Mar 22 11:40:35 ppubssa3ed setroubleshoot[1544]: SELinux is preventing /usr/sbin/nginx from name_bind access on the tcp_ socket port 89. For complete SELinux messages run: sealert -l f2914dba-04ef-44ca-9a0b-0f5e62ec72e4

We can look at a complete explanation through the sealert command (as mentioned in the log), as follows: # sealert -l f2914dba-04ef-44ca-9a0b-0f5e62ec72e4 SELinux is preventing /usr/sbin/nginx from name_bind access on the tcp_socket port 89. ***** Plugin bind_ports (99.5 confidence) suggests ************************ If you want to allow /usr/sbin/nginx to bind to network port 89 Then you need to modify the port type. Do

74

Understanding SELinux Decisions and Logging

# semanage port -a -t PORT_TYPE -p tcp 89 where PORT_TYPE is one of the following: http_cache_port_t, http_port_t, jboss_management_port_t, jboss_messaging_port_t, ntop_port_t, puppet_port_t. ***** Plugin catchall (1.49 confidence) suggests ************************** ...

The sealert application is a command-line application that parses the information stored by the setroubleshoot daemon (in /var/lib/setroubleshoot). This will provide us with a set of options to resolve the denial. In the case of the Apache-related denial shown earlier, sealert gives us one option with a certain confidence score. Depending on the problem, this tool might show multiple options, each with its own confidence figure (that is, how certain sealert is that this is the right resolution). As we can see from this example, the setroubleshoot application itself uses plugins to analyze denials. These plugins (offered through the setroubleshoot-plugins package) look at a denial to check whether they match a particular, well-known use case (for example, when to change an SELinux Boolean or when a target resource has a wrong context) and give feedback to setroubleshoot about how certain the plugin is so that this denial can be resolved through its recommended method.

Sending emails when SELinux denials occur Once a system is fine-tuned and denials no longer occur regularly, administrators can opt to have setroubleshootd send emails whenever a new denial comes up. This truly brings SELinux's host intrusion detection/prevention capabilities on top, as administrators do not need to constantly watch their logs for information. However, keep in mind that this could lead to a sudden burst in emails, which might result in Denial of Service (DoS)-like behavior, if many denials are triggered. Administrators should only implement this if their email infrastructure has rate limiting or other Quality of Service (QoS) controls in place.

Getting help with denials

75

Open /etc/setroubleshoot/setroubleshoot.conf in a text editor and locate the [email] section. Update the parameters to match the local mailing infrastructure, as follows: # vi /etc/setroubleshoot/setroubleshoot.conf [email] recipients_filepath = /var/lib/setroubleshoot/email_alert_ recipients smtp_port = 25 smtp_host = localhost from_address = [email protected] subject = [infra] SELinux Alert for host infra.example.com

Next, edit the email_alert_recipients file (as referenced through the recipients_filepath variable), and add the email addresses that need to be notified when an SELinux alert comes up. Finally, restart the D-Bus daemon, as follows: # systemctl restart dbus

When working on a non-systemd system, use the following command instead: # service dbus restart

The D-Bus restart is needed as D-Bus manages the setroubleshootd daemon.

Using audit2why If setroubleshoot and sealert are not available in the Linux distribution, we can still get some information about a denial. Although it isn't as extensive as the plugins offered by setroubleshoot, the audit2why utility (which is short for audit2allow -w) does provide some feedback on a denial. Sadly, it isn't always right in its deduction. Let's try it out against the same denial for which we used sealert, as follows: # ausearch -m avc -ts recent | audit2why type=AVC msg=audit(1584880436.644:385): avc: denied { name_bind } for pid=5119 comm="nginx" src=89 scontext=system_u:system_r:httpd_t:s0 tcontext=system_u:object_r:reserved_port_t:s0 tclass=tcp_socket permissive=0 Was caused by:

76

Understanding SELinux Decisions and Logging Missing type enforcement (TE) allow rule.

You can use audit2allow to generate a loadable module to allow this access.

The audit2why utility here didn't consider that the context of the target location was wrong, and it suggests that the policy be updated to allow the web server to bind to the unreserved_port_t type, unlike the information provided by setroubleshoot, which was more accurate, recommending that the target port be relabeled instead. As the output of the command mentions, another tool exists called audit2allow, which can convert a denial into an SELinux policy. We will cover audit2allow in Chapter 12, Tuning SELinux Policies.

Interacting with systemd-journal Alongside the Linux audit system, which is used for most SELinux logging and events, we can also gather information through other logging systems. systemd's journal, for instance, captures SELinux context information with the events and allows administrators to use this information while querying the journal. For instance, to see the events in systemd-journal that are generated by an application associated with the system_u:system_r:sssd_t:s0 context, the following command can be used: # journalctl _SELINUX_CONTEXT="system_u:system_r:sssd_t:s0" -- Logs begin at Sun 2020-03-22 10:43:48 UTC, end at Sun 202003-22 12:40:12 UTC. -Mar 22 10:43:51 ppubssa3ed sssd[545]: Starting up Mar 22 10:43:51 ppubssa3ed sssd[be[implicit_files]][623]: Starting up Mar 22 10:43:51 ppubssa3ed sssd[nss][630]: Starting up

Because systemd-journal adds the SELinux context of the originating application, it is harder for malicious applications to generate fake events. Whereas regular system loggers just capture string events, systemd-journal retrieves the SELinux context from the system. Using the SELinux context, it is easy to group events across different but strongly related applications and have a higher guarantee that events come from a particular application.

Getting help with denials

77

When the bash-completion package is installed, we can even use it to see which SELinux contexts are present in the systemd-journal logs, which makes querying the journal logs much easier, as follows: # journalctl _SELINUX_CONTEXT= kernel system_u:system_r:auditd_t:s0 system_u:system_r:chronyd_t:s0 ...

To find messages related to nginx, use the embedded grep filter, as follows: # journalctl -g nginx -- Logs begin at Sun 2020-03-22 10:43:48 UTC, end at Sun 202003-22 12:52:26 UTC. -Mar 22 11:40:32 ppubssa3ed systemd[1]: Starting The nginx HTTP and reverse proxy server... Mar 22 11:40:32 ppubssa3ed nginx[1538]: nginx: the configuration file /etc/nginx/nginx.conf syntax is ok Mar 22 11:40:32 ppubssa3ed nginx[1538]: nginx: [emerg] bind() to 0.0.0.0:89 failed (13: Permission denied) ... Mar 22 11:40:35 ppubssa3ed setroubleshoot[1544]: SELinux is preventing /usr/sbin/nginx from name_bind access on the tcp_ socket port 89. For complete SELinux messages run: sealert -l f2914dba-04ef-44ca-9a0b-0f5e62ec72e4

The benefit of the embedded grep filter is that journalctl will still show the multiline messages, whereas actually redirecting the journalctl output through grep would only show the individual lines that match the expression.

Using common sense Common sense is not easy to document, but reading a denial often leads to the right solution when we have some experience with file labels (and what they are used for). If we get a denial about a web server failing to read its files, and the context of the file is (for instance) user_home_t, then that should ring a bell. End user home files, for instance, use the user_home_t context, which is not suitable for system files that the web server reads.

78

Understanding SELinux Decisions and Logging

One way to make sure that the context of the target resource is correct is to verify it with matchpathcon. This utility returns the context as it should be according to the SELinux policy, as follows: $ matchpathcon /srv/www/html/index.html /srv/www/html/index.html system_u:object_r:httpd_sys_ content_t:s0

Performing this for denials related to files and directories might help in finding a proper solution quickly. Furthermore, many domains have specific manual pages that inform the reader about types commonly used for each domain, as well as how to deal with the domain in more detail (for example, the available booleans, common mistakes made, and so on). These manual pages start with the main service and are suffixed with _selinux, as illustrated here: $ man ftpd_selinux

In most cases, the approach to handling denials can be best described as follows: • Is the target resource label (such as the file label) the right one? Verify this with matchpathcon, or compare with labels of working (accessible) resources. • Is the source label (the domain) the expected one? An SSH daemon should run in the sshd_t domain, not the init_t domain. If this is not the case, make sure that the labels of the application itself (such as its executable binary) are correct (again, use matchpathcon for this). • Is the denial one that might be covered by an SELinux boolean? In that case, the policy might already have the appropriate rules in place, only requiring a change in an SELinux boolean value. setroubleshootd will report this if it is the case. Usually, the manual page of the domain (such as httpd_selinux) will also cover the available SELinux Booleans. We explain how to query and adjust SELinux Booleans in Chapter 12, Tuning SELinux Policies. Changing file labels will be discussed in Chapter 4, Using File Contexts and Process Domains. To close off this section, common sense will be your most prolific approach to managing SELinux denials, but the aforementioned tools will be of assistance to begin with.

Summary

79

Summary In this chapter, we saw how to enable and disable SELinux, both on a complete system level as well as a per-service level using various methods: kernel boot options, an SELinux configuration file, or plain commands. One such command is semanage permissive, which can disable SELinux protections for a single service. Next, we saw where SELinux logs its events and how to interpret them, which is one of the most common tasks an administrator has to undertake when dealing with SELinux. To assist us with this interpretation, we can use tools such as setroubleshoot, sealert, and audit2why. We also touched upon several utilities related to Linux auditing to help us sift through various events. In the next chapter, we will look at the first administrative task on SELinux systems: managing user accounts, their associated SELinux roles, and security clearances for the resources on the system.

Questions 1. What should administrators try before disabling SELinux? 2. Where can administrators find SELinux logs by default? 3. How do we know whether an application is SELinux-aware? 4. What is the purpose of the AVC? 5. Are AVC events the only type of events for SELinux?

3

Managing User Logins When we log in to an SELinux-enabled system, we receive an SELinux context to work in. This context contains an SELinux user, an SELinux role, a domain, and optionally, a sensitivity range. As the SELinux user defines the roles and types that can be accessed, managing user logins and SELinux users is the first step in configuring end users on the system. To enable properly configured users, we will learn to define users that have sufficient rights to do their jobs, ranging from regular users with strict SELinux protections to fully privileged administrative users with few SELinux protections. We will create and assign categories and sensitivities, as well as assign roles to users and use various tools to switch roles. At the end of the chapter, we will see how SELinux integrates with the Linux authentication process. In this chapter, we're going to cover the following main topics: • User-oriented SELinux contexts • SELinux users and roles • Handling SELinux roles • SELinux and PAM

82

Managing User Logins

Technical requirements Check out the following video to see the Code in Action: https://bit.ly/3jbASmr

User-oriented SELinux contexts Once logged in to a system, our user will run inside a certain context. This user context defines the rights and privileges that we, as a user, have on the system. The command to obtain current user information, id, also supports displaying the current SELinux context information: $ id -Z unconfined_u:unconfined_r:unconfined_t:s0-s0:c0.c1023

On SELinux systems with a targeted policy type, chances are very high that all users are logged in as unconfined_u (the first part of the context). On more restricted systems, the user can be user_u (regular restricted users), staff_u (operators), sysadm_u (system administrators), or any of the other SELinux users. The SELinux user defines the roles that the user can switch to. SELinux roles themselves define the application domains that the user can use. By default, a fixed number of SELinux users are available on the system, but administrators can create additional SELinux users. It is also the administrator's task to assign Linux logins to SELinux users. SELinux roles, on the other hand, cannot be created through administrative commands, as SELinux roles are part of the SELinux policy. For this, the SELinux policy needs to be enhanced with additional rules that create the role. We will touch upon that in Chapter 15, Using the Reference Policy. To view the currently available roles, use seinfo: # seinfo --role Roles: 14 auditadm_r dbadm_r ... xguest_r

SELinux roles can be coarse-grained (such as sysadm_r) or more functionality-oriented (such as dbadm_r). Custom SELinux roles can even be very fine-grained, only granting the ability to transition into limited domains.

SELinux users and roles

83

Let's see how to create and manage SELinux users.

SELinux users and roles In SELinux-enabled environments, the login binary calls the libselinux API to establish the initial mapping between SELinux users and local users. Then, after finding the right SELinux user, the system looks up the role and domain that the user should be in and sets that as the user's context.

Listing SELinux user mappings When logged in to the system, we can use id -Z to obtain the current SELinux context. For many users, this context will be defined by the unconfined user (unconfined_u), regardless of their username. If not that, it will generally be a context based on one of sysadm_u, staff_u, or user_u. This is because most Linux distributions will only provide a limited set of SELinux users by default, aligned with the SELinux roles that they support. During login, the service process through which the login is handled will check a local definition file to find the appropriate mapping between the Linux account and the SELinux user. Let's look at the existing login mappings using semanage login -l. The following output is the default output on a CentOS system: # semanage login -l Login Name SELinux User

MLS/MCS Range

Service

__default__ unconfined_u s0-s0:c0.c1023 * root unconfined_u s0-s0:c0.c1023 *

The output of the command shows one login mapping per line. Each mapping consists of the following: • The Login Name for which the mapping is applicable (that is, the username) • The SELinux User to which the login is mapped • The MLS/MCS Range to which the login is mapped • The Service for which the mapping applies (this is used for local customizations, which we will tackle in the Customizing logins for services section)

84

Managing User Logins

The login name can contain a few special values that do not map directly to a single Linux account: • __default__ is a catch-all rule. If none of the other rules match, then the users are mapped to the SELinux user identified through this line. In the given example, all users are mapped to the unconfined_u SELinux user, meaning regular Linux users are hardly confined in their actions. When this isn't meant to happen, administrators usually map regular logins to restricted SELinux users, while administrative logins are mapped to the staff_u or sysadm_u SELinux users. • Login names starting with % will map to groups. This allows administrators to map a group of people directly to an SELinux user rather than having to manage the mappings individually. When both an individual user mapping and group mapping match, then the individual user mapping takes precedence. When multiple group definitions exist, then SELinux will use the first matching group mapping (in the order listed in the underlying seusers configuration file). Important note System processes (non-interactively logged-in Linux accounts) are mapped to the system_u SELinux user. This SELinux user should never be assigned to end user logins.

In the case of an MLS- or MCS-enabled system, the mapping contains information about the user's allowed sensitivity range (MLS/MCS range). This way, we can map multiple users to the same restricted SELinux user, while differentiating between these users through the allowed sensitivities. For instance, one user might only be allowed to access low-sensitivity areas (s0), whereas another user might also have access to higher sensitivities (for example, s1) or different categories.

Mapping logins to SELinux users Let's use a few examples to show how we make these mappings work. For more intricate details on this, see the SELinux and PAM section. We'll assume we have a Linux user called lisa, and we want her account to be mapped to the staff_u SELinux user, whereas all other users in the users group are mapped to the user_u SELinux user.

SELinux users and roles

85

We can accomplish this through the semanage login command, using the -a (add) option: # semanage login -a -s staff_u lisa # semanage login -a -s user_u %users

The -s parameter assigns the SELinux user to the given login, whereas the -r parameter handles the sensitivity (and categories) for that user. For instance, let's modify (using -m instead of -a) the recently created group-based definition by mapping to the staff_u user instead, and limiting these users to the s0-s0 sensitivity range and categories c0 to c4: # semanage login -m -s staff_u -r "s0-s0:c0.c4" %users

The sensitivity range of a login mapping may not exceed the range assigned to the SELinux user. For example, if the staff_u SELinux user itself is only granted access to s0-s0:c0.c3, then the previous command will fail as it is trying to assign a broader access range. We'll discuss how to define SELinux users and their range in the Creating SELinux users section. The semanage login command updates the seusers file located inside /etc/ selinux/targeted. If multiple group mappings are defined, then the order of mappings within this file defines which mapping applies to a given user. Users that belong to multiple mapped groups will be assigned an SELinux user based on the first match. While it is possible to update the order of the entries in the seusers file, this is not recommended. Every time semanage login modifies the seusers file, it will reorder the mappings. Instead, when a user belongs to multiple mapped groups, we advise you to create an individual (user-based) mapping. This is also shown whenever we create a group mapping for a group that contains users for which active mappings already exist: # semanage login -a -s guest_u %nginx libsemanage.add_user: User taylor is already mapped to group users, but also belongs to group nginx. Add an explicit mapping for this user to override group mappings.

The changes take effect when a new login occurs, so we should force a logout for these users. The following command kills all the processes of the lisa user, forcing a logout for that user: # pkill -KILL -u lisa

86

Managing User Logins

Also, when we modify a user's settings, we should also reset the contexts of that user's home directory (while that user is not logged in). To accomplish this, use restorecon as follows: # restorecon -RF /home/lisa

The -F option in the preceding command forces a reset, while -R does this recursively. Important note Running the restorecon -RF command will also reset file contexts that the user has manually set using tools such as chcon. We recommend defining SELinux user mappings up front, or recursively changing only the SELinux user of the files using chcon -R -u. The chcon application and file contexts are discussed in the next chapter.

To remove a login mapping, use the -d (delete) option. Don't forget to run the restorecon command afterward: # semanage login -d lisa # restorecon -RF /home/lisa

Don't forget to force a user logout again if this user is active on the system.

Customizing logins for services When login mappings are added using semanage login, they apply to all services. There is no option in semanage to allow customizing the mappings based on the service. However, that does not mean it is not possible. The SELinux user space tools and libraries will consult the following two configuration files to know what the mappings are: • The /etc/selinux/targeted/seusers file contains the standard, service-agnostic mappings. This file is managed by semanage login and should not be updated through any other means. • The /etc/selinux/targeted/logins directory contains customized mappings, one file per Linux account. So, the custom mapping for the root user will be in /etc/selinux/targeted/logins/root.

SELinux users and roles

87

Inside the files for customized mappings, administrators can define, per service, a different SELinux user to map to. The services are the Pluggable Authentication Modules (PAM) services through which a user can log in, and more information on this can be found in the SELinux and PAM section. For instance, to have the root user – when logged in through SSH – be mapped to the user_u SELinux user rather than their default unconfined_u user, the root file would need to contain the following: sshd:user_u:s0

When querying the current mapping, semanage login will show this customization as follows: # semanage login -l ... Local customization in /etc/selinux/targeted/logins root user_u s0 sshd

Of course, this customization does not need to be so drastic. It can also be used to limit the user's default MLS/MCS range. For instance, to limit the categories to c0.c8 (rather than the default c0.c1023 range) you would use the following: sshd:unconfined_u:s0-s0:c0.c8

Such customizations allow us to flexibly change the access control policies based on the PAM service used.

Creating SELinux users By default, only a small number of SELinux users are available for mapping to logins. If we want more control over the Linux accounts and their mappings, we need to create additional SELinux users. First, list the currently known SELinux users using the semanage user -l command, as follows: # semanage user -l SELinux Labeling MLS/ MLS/ User Prefix MCS Level MCS Range SELinux Roles guest_u user s0 s0 guest_r root user s0 s0-s0:c0.c1023 staff_r

88

Managing User Logins

... ... xguest_u user s0 s0 xguest_r

Next, create a new SELinux user with semanage user, using the -a (add) option. We need to give SELinux additional information about this SELinux user, such as the following: • The default sensitivity (using the -L option) for the SELinux user. This is the sensitivity that the user starts with. • The security clearance (using the -r option) applicable to the SELinux user. This range cannot be extended when defining login mappings. It is, however, possible to give a user a more limited range, as long as it is bounded by the current range. • The allowed role or roles (using the -R option) for the SELinux user. Tip The labeling prefix shown in the previous example is used to dynamically create SELinux policies with specific prefixes, such as _home_t for user's home files. Most distributions leave this to the default user setting, and changing it is done through the (undocumented) -P parameter to semanage user.

In the following example, we're configuring the SELinux user finance_u: # semanage user -a -L s0 -r "s0-s0:c0.c127" -R user_r finance_u

When the command creates the SELinux user, its information becomes part of the SELinux policy. From this point onward, administrators can map Linux accounts to this SELinux user. Important note SELinux roles are enabled through the SELinux user that a Linux account is mapped to. When an administrator wants to allow additional existing roles to a Linux account, the administrator either updates existing SELinux mappings to include the new role(s) or creates a new SELinux user that has access to the new role(s) and then maps this SELinux user to the Linux account.

SELinux users and roles

89

Just like with login mappings, semanage user also accepts the -m option to modify an existing entry, or -d to delete one. For instance, the following command deletes the finance_u SELinux user: # semanage user -d finance_u

Separate SELinux users enhance the audit information since SELinux users generally do not change during a user's session, whereas the effective Linux user ID can. If the user creates files or other resources, these resources also inherit the SELinux user part in their security context.

Listing accessible domains When creating SELinux users, one of the parameters that needs to be provided is the role or roles for an SELinux user. Most of the roles are self-explanatory: the dbadm_r role is for DBAs, whereas the webadm_r role is for web application infrastructure administrators. If a role is not clear, or an administrator is not certain which accesses are part of a given role, the administrator can still query the SELinux policy for more information. Informational note This book will mostly focus on the command-line utilities used to query and interact with the active SELinux policy. In Chapter 13, Analyzing Policy Behavior, we will also cover the graphical utility apol.

As documented earlier, roles define which domains are accessible for the users associated with the role. We saw that seinfo can show us the available roles, but it can do more. It can list the domains accessible for a role as well, using the -x option: # seinfo -r dbadm_r -x Roles: 1 role dbadm_r types { ... qmail_inject_t user_mail_t ... };

In this example, users running with the dbadm_r role as part of their security context will be able to transition to, for instance, the qmail_inject_t (the domain used to read email messages and pass those on to the qmail queue) and user_mail_t (the domain used for generic email-sending command-line applications) domains. The information provided through the dominated roles is usually not of concern to the administrators. Role dominance, although supported in SELinux core, is not used by Linux distribution policies. It signifies the inheritance of (other) roles, but it will always just show the queried role.

90

Managing User Logins

Managing categories Sensitivity labels and their associated categories are identified through numeric values, which is great for computers but not that obvious for users. Luckily, the SELinux utilities support translating the levels and categories to human-readable values, even though they are still stored as numbers. As a result, almost all tools that can show contexts will show them translated rather than presented as numerical values. The translations are managed through the setrans.conf file, located in /etc/ selinux/targeted. Inside this file, we can name specific values (for example, s0:c102) or ranges (such as s0-s0:c1.c127) with a string that is much easier for administrators to use. However, for translations to be performed, mcstransd – the MCS translation daemon – needs to be running. Consider our example of the finance_u SELinux user who was allowed access to the c0.c127 category range. Two of the categories within that range are c102, which we will tag as Contracts, and c103, which we will tag as Salaries. The c1.c127 range will be labeled as FinanceData. The following diagram shows the relationship between these various categories:

Figure 3.1 – Relationship of the example categories and category range

To accomplish this, the following should be placed in the setrans.conf file: s0:c102=Contracts s0:c103=Salaries s0-s0:c1.c127=FinanceData

After editing the setrans.conf file, the mcstransd application needs to be restarted. These translations are handled by the SELinux utilities, which connect to the mcstransd daemon through the .setrans-unix socket located in /var/run/setrans to query the setrans.conf file. If the daemon is not running or the communication with the daemon fails, the numeric sensitivity and category values are displayed.

Handling SELinux roles

91

For instance, with the daemon running, the output of id -Z is now as follows: # id -Z unconfined_u:unconfined_r:unconfined_t:SystemLow-SystemHigh

We can view the available sensitivities and their human-readable counterparts using the chcat tool. The following example displays the translations after adding the financerelated ones: $ chcat -L s0 SystemLow s0-s0:c0.c1023 SystemLow-SystemHigh s0:c0.c1023 SystemHigh s0:c102 Contracts s0:c103 Salaries s0-s0:c1.c127 FinanceData

The same chcat utility can be used to assign categories to users. For instance, to grant the Salaries category to the lisa Linux user, we'd use the following command: # chcat -l -- +Salaries lisa

The previous command grants the Salaries category (c103) to the Linux user lisa. The user mapping is immediately updated with this information. Again, we need to make sure that the lisa user is logged out for the changes to take effect. With this, we end our section on managing SELinux users and logins. We've learned how to align users with SELinux users so that they log in to the system with the correct context. In the next section, we will look at the SELinux roles and how to apply those to SELinux users.

Handling SELinux roles We saw how SELinux users define the role(s) that a user can hold. But how does SELinux enforce which role a user logs in through? And when logged in, how can a user switch their active role?

Defining allowed SELinux contexts To select the context assigned to a successfully authenticated user, SELinux introduces the notion of a default context. Based on the context of the service through which a user logs in (or through which the user executes commands), the system selects the right user context.

92

Managing User Logins

Inside the /etc/selinux/targeted/contexts directory, a file called default_contexts exists. Each line in this file starts with the SELinux context information of the parent process and is then followed by an ordered list of all the contexts that could be picked based on the user's allowed SELinux role(s). Consider the following line of code for the sshd_t context: system_r:sshd_t:s0

user_r:user_t:s0 \ staff_r:staff_t:s0 \ sysadm_r:sysadm_t:s0 \ unconfined_r:unconfined_t:s0

This line of code mentions that when a user logs in through a process running in the sshd_t domain, the listed roles are checked against the roles of the user. The user will transition to the first context listed that matches the roles the user can use. For instance, assume we are mapped to an SELinux user that has access to both the staff_r and sysadm_r roles. In that case, we will log in as staff_r:staff_t since that is the first match. However, like the seusers file for the Linux account mappings, the default_ contexts file is a default file that can be overruled through specific customizations. These customizations are stored in the /etc/selinux/targeted/contexts/ users subdirectory. These files are named after the SELinux user for which they take effect. This allows us to assign different contexts for particular SELinux users even if they share the same roles with other SELinux users. Because SELinux checks the entries per line, we do not need to copy the entire content of the default_contexts file. Only the configuration lines that we want to see a different configuration for need to be listed; SELinux will automatically use the default_contexts file for the rest. Let's modify the default contexts so that the staff_u SELinux user logs in with the sysadm_r role (and with the sysadm_t type) when logged in through SSH. To do so, use the sshd_t line, modify it, and save the result as /etc/selinux/targeted/ contexts/users/staff_u: system_r:sshd_t:s0 sysadm_r:sysadm_t:s0

Specifically, for the SSH daemon, we also need to enable the ssh_sysadm_login boolean, which is a special precaution SELinux policy developers have made to prevent users from immediately logging in with highly privileged accounts: # setsebool ssh_sysadm_login on

Handling SELinux roles

93

With these settings in place, we've set sysadm_r:sysadm_t:s0 as the only possible context, ensuring that the target context is staff_u:sysadm_r:sysadm_t.

Validating contexts with getseuser To validate whether our change succeeded, we can ask SELinux what the result of a context choice will be without having to parse the files ourselves. We can accomplish this through the getseuser command, which takes two arguments: the Linux user account and the context of the process that switches the user context. Important note The getseuser command is a helper utility offered by the SELinux user space project, but is not made available on all distributions. You will find it on Debian and Gentoo, but not on CentOS or other Red Hat Enterprise Linuxderived distributions.

Here's an example that checks what the context would be for the sven user when they log in through a process running in the sshd_t domain: # getseuser sven system_u:system_r:sshd_t seuser: user_u, level s0-s0 Context 0 user_u:user_r:user_t:s0

One of the advantages of the getseuser command is that it asks the SELinux code what the context should be, which not only looks through the default_contexts and customized files, but also checks whether the target context can be reached or not, and that there are no other constraints that prohibit the change to this context.

Switching roles with newrole After having successfully authenticated and logged in, users will be assigned the context through the configuration mentioned in the SELinux users and roles section. If the SELinux user has access to multiple roles, however, then the Linux user can use the newrole application to transition from one role to another.

94

Managing User Logins

Consider an SELinux system without unconfined domains and where we are, by default, logged in as the staff_r role. To perform administrative tasks, we need to switch to the sysadm_r administrative role, which we can do with the newrole command. This command only works when working through a secure terminal listed in /etc/ securetty: $ id -Z staff_u:staff_r:staff_t:s0 $ newrole -r sysadm_r Password: (Enter user password) $ id -Z staff_u:sysadm_r:sysadm_t:s0

Notice how the SELinux user remains constant but the role and domain have changed. The newrole command can also be used to transition to a specific sensitivity, as follows: $ newrole -l s0-s0:c0.c100

When we switch to another role or sensitivity, what we actually do is create a new session (with a new shell) that has this new role or sensitivity. The command does not change the context of the current session, nor does it exit from the current session. We can return from our assigned role and go back to the first session by exiting (through exit, logout, or Ctrl + D).

Managing role access through sudo Most administrators use sudo for privilege delegation: allowing users to run certain commands in a more privileged context than the user is otherwise allowed. The sudo application is also capable of switching SELinux roles and types. We can pass the target role and type to sudo directly. For instance, we can tell sudo to switch to the administrative role when we edit a PostgreSQL configuration file: $ sudo -r sysadm_r -t sysadm_t vim /var/lib/pgsql/data/pg_hba. conf

Handling SELinux roles

95

However, we can also configure sudo through the /etc/sudoers file to allow users to run commands within a certain role and/or type, or get a shell within a certain context. Consider a user that has access to both the user_r and dbadm_r roles (with the dbadm_r role being a role designated for database administrators). Within the sudoers file, the following line allows the myuser user to run any command through sudo, which, when triggered, will run with the dbadm_r role and within the dbadm_t domain: myuser ALL=(ALL) TYPE=sysadm_t ROLE=sysadm_r ALL

Often, administrators will prefer sudo over newrole as the latter does not change the effective user ID, which is often required for end users when they want to invoke a more privileged command (concerning the root user or a service-specific runtime account) anyway. The sudo application also has great logging capabilities, and we can even have commands switching roles without requiring the end user to explicitly mention the target role and type. Sadly, it does not support changing sensitivities.

Reaching other domains using runcon Another application that can switch roles and sensitivities is the runcon application. The runcon command is available for all users and is used to launch a specific command as a different role, type, and/or sensitivity. It even supports changing the SELinux user – assuming the SELinux policy lets you. The runcon command does not have its own domain – it runs in the context of the user executing the command. As such, the privileges of the user domain itself govern the ability to change the role, type, sensitivity, or even SELinux user. Most of the time, we will use runcon to launch applications with a particular category. This allows us to take advantage of the MCS approach in SELinux without requiring applications to be MCS-enabled: $ runcon -l Salaries bash $ id -Z unconfined_u:unconfined_r:unconfined_t:Salaries

For instance, in the previous example, we run a shell session with the Salaries category (prohibiting it from accessing resources that do not have the same or fewer categories set).

96

Managing User Logins

Switching to the system role Sometimes, administrators will need to invoke applications that should not run under their current SELinux user context but instead as the system_u SELinux user with the system_r SELinux role. SELinux policy administrators acknowledge this need, and allow a very limited set of domains to switch the SELinux user to a different user – perhaps contrary to the purpose of the immutability of SELinux users mentioned earlier. Yet, as there are cases where this is needed, SELinux needs to accommodate this. One of the applications allowed to switch the SELinux user is run_init (through its run_init_t domain). The run_init application is mainly (almost exclusively) used to start background system services on a Linux system. Using this application, the daemons do not run under the user's SELinux context but the system's, as required by SELinux policies. As this is only needed on systems where launching additional services is done through service scripts, distributions that use systemd do not require the use of run_init. systemd already runs with the system_r role and is responsible for starting additional services. As such, no role transition is needed. Other init systems, such as Gentoo's OpenRC, integrate run_init so that administrators do not generally need to invoke run_init manually. Most SELinux policies enable role-managed support for selective service management (for non systemd distributions). This allows users that do not have complete system administration rights to still manipulate a select number of services on a Linux system, allowed by the SELinux policy. These users are to be granted the system_r role, but once accomplished, they do not need to call run_init to manipulate specific services anymore. The transitions happen automatically and only for the services assigned to the user – other services cannot be launched by these users. This finalizes our section on handling SELinux roles. We've learned how to manage SELinux roles and switching roles and contexts, as well as how to define the target role and type in the case of privilege escalation. In the last section of this chapter, we will look at how PAM is used to configure the SELinux context setup on the system.

SELinux and PAM With all the information about SELinux users and roles, we have not touched upon how exactly applications or services create and assign an SELinux context to a user. As mentioned earlier on, this is coordinated through the use of Linux's PAM services.

SELinux and PAM

97

Assigning contexts through PAM End users log in to a Linux system through either a login process (triggered through a getty process), a networked service (for example, the OpenSSH daemon), or through a graphical login manager (xdm, kdm, gdm, slim, and so on). These services are responsible for switching our effective user ID (upon successful authentication, of course) so that we are not active on the system as the root user. For SELinux systems, these processes also need to switch the SELinux user (and role) accordingly, as otherwise, the context will be inherited from the service, which is obviously wrong for any interactive session. In theory, all these applications can be made fully SELinux aware, linking with the SELinux user space libraries to get information about Linux mappings and SELinux users. Instead of converting all these applications, the developers decided to take the authentication route to the next level using the PAM services that Linux systems provide. PAM offers a very flexible interface for handling different authentication methods on Linux (and Unix) systems. All applications mentioned earlier use PAM for their authentication steps. To enable SELinux support for these applications, we need to update their PAM configuration files to include the pam_selinux.so library. The following code listing is an excerpt from CentOS's /etc/pam.d/remote file, limited to PAM's session service directives. It triggers the pam_selinux.so library code as part of the authentication process, as follows: session required session required session required session required session optional session include session include

pam_selinux.so close pam_loginuid.so pam_selinux.so open pam_namespace.so pam_keyinit.so force revoke password-auth postlogin

The arguments supported by the pam_selinux.so code are described in the pam_ selinux manual page. In the preceding example, the close option clears the current context (if any), whereas the open option sets the context of the user. The pam_selinux module takes care of querying the SELinux configuration and finding the right mappings and context based on the service name used by the daemon.

98

Managing User Logins

Prohibiting access during permissive mode Having SELinux active and enforcing on a system improves its resilience against successful exploits and other malicious activities, especially when the system is used as a shell server (or provides other interactive services) and the users are confined – meaning they are mapped to user_u or other confined SELinux users. Some administrators might want to temporarily switch the system to permissive mode. This could be to troubleshoot issues or to support some changes made on the system. When using permissive mode, it would be a good idea to ensure that the interactive services are not usable for regular users. With pam_sepermit, this can be enforced on the system. The PAM module will deny a set of defined users access to the system if the system is in permissive mode. By default, these users are mentioned in /etc/security/sepermit.conf, but a different file can be configured through the conf= option inside the PAM configuration itself. In the sepermit.conf file, there are three approaches to document which users should be denied access when the system is in permissive mode: • Regular usernames • Group names, prefixed with the @ sign • SELinux usernames, prefixed with the % sign Within this file, we list each user, group, or SELinux user on a single line. After each entry, we can (but don't have to) add one or two options: • exclusive means that the system will allow the user to be active even when the system is in permissive mode, but only a single session can be active. When the user logs out, all active processes will be killed. • ignore will return PAM_IGNORE as the return status if SELinux is in enforcing mode, and PAM_AUTH_ERR if SELinux is in permissive mode. This allows special constructs/branches for this user in PAM based on the permissive state of the system. To enable pam_sepermit, it's sufficient to enable the module in the auth PAM service as follows: auth required

pam_sepermit.so

Of course, don't forget to remove all active user sessions when switching to permissive mode, as any running session is otherwise left untouched.

SELinux and PAM

99

Polyinstantiating directories The last PAM module we'll look at is pam_namespace.so. Before we dive into configuring this module, let's first look at what polyinstantiation is about. Polyinstantiation is an approach where, when a user logs in to a system, the user gets a view on filesystem resources specific to its session, while optionally hiding the resources of other users. This differs from regular access controls, where the other resources are still visible, but might just be inaccessible. This session-specific view, however, does not just use regular mounts. The module uses the Linux kernel namespace technology to force a (potentially more limited) view on the filesystem, isolated and specific to the user session. Other users have a different view on the filesystem. Let's use a common example. Assume that all users, except root, should not have access to the temporary files generated by other users. With standard access controls, these resources would still be visible (perhaps not readable, but their existence or the directories they reside in would be visible). Instead, with polyinstantiation, a user will only see their own /tmp and /var/tmp views. The following setting in /etc/security/namespace.conf will remap these two locations: /tmp /tmp/tmp-inst/ level root /var/tmp /var/tmp/tmp-inst/ level root

On the real filesystem, those locations will be remapped to a subdirectory inside /tmp/ tmp-inst and /var/tmp/tmp-inst. The end users do not know or see the remapped locations – for them, /tmp and /var/tmp are as they would expect. The format (and thus polyinstantiation) of the subdirectories created depends on the third option within the namespace.conf file. The supported options are as follows: • user, which will create a subdirectory named after the user (such as lisa) • level, which will create a subdirectory named after the user sensitivity level and username (such as system_u:object_r:tmp_t:s0-s0:c0.c1023_lisa) • context, which will create a subdirectory named after the process context (including the sensitivity level) and username (such as system_u:object_r:user_tmp_t:s0_lisa)

100

Managing User Logins

For SELinux systems, the most common setting is level. Tip In the default namespace.conf file, you might notice that this also has support for home directories. When enabled with the level or context method, this will ensure that users have a sensitivity-specific home directory set. For instance, if the system configuration forces the user to have a lower sensitivity when logged in through SSH than when the user logs in through the terminal, a different home directory view will be used.

In the previous example, only the root user is exempt from these namespace changes. Additional users can be listed (comma-separated), or an explicit list of users can be given for which polyinstantiation needs to be enabled (if we prefix the user list with the ~ character). To allow the namespace changes to take place, the target locations need to be available on the system with the 000 permission: # mkdir /tmp-inst && chmod 000 /tmp-inst

Next, enable pam_namespace.so in the PAM configuration files at the session service: session required pam_namespace.so

Finally, make sure that SELinux allows polyinstantiated directories. On CentOS, this is governed through the polyinstantiation_enabled SELinux boolean: # setsebool polyinstantiation_enabled on

Other distributions will have it through the allow_polyinstantiation SELinux boolean. With the polyinstantiation support, we close off this final section of the chapter, where we learned how PAM is used to trigger SELinux context changes on the system.

Summary

101

Summary SELinux maps Linux users onto SELinux users and defines the roles a user can be assigned through the SELinux user definitions. We learned how to manage those mappings and SELinux users with the semanage application, and how to grant the right roles to the right people. We also saw how the same commands are used to grant the proper sensitivity to the user and how we can describe these levels in the setrans.conf file. We used the chcat tool to do most of the category-related management activities. After assigning roles to the users, we saw how to jump from one role to another using newrole, sudo, runcon, and run_init. We ended this chapter with important insights into how SELinux is integrated into the Linux authentication process and how to tune a Linux system further using a couple of SELinux-aware PAM modules. In the next chapter, we will learn to manage the labels on files and processes, and see how we can query the SELinux policy rules.

Questions 1. Why can't we just add an SELinux role to a Linux account? 2. Can Linux accounts be mapped to more than one SELinux user? 3. Besides associating the valid SELinux roles, what other advantages does an SELinux user have? 4. What purpose does PAM have when dealing with Linux accounts and SELinux mappings?

4

Using File Contexts and Process Domains SELinux-enabled systems are strongly dependent on the notion of contexts (on resources) and domains (on processes). The access controls that SELinux enforces use these contexts to identify the resources, and define the enforcement rules within the policy. Because of its inherent reliance on these contexts, this chapter will go into detail on file contexts, context definitions, and process domains. We will work with the file contexts and learn where they are stored so that you can easily adjust your system to work optimally with SELinux. We assign contexts to resources both temporarily (for testing purposes) and permanently, and learn how these contexts are used to automatically deduce the process domain. Once we know how to obtain process domain information, we will query the SELinux policy to learn about the current access controls.

104

Using File Contexts and Process Domains

In this chapter, we're going to cover the following main topics: • Introduction to SELinux file contexts • Keeping or ignoring contexts • SELinux file context expressions • Modifying file contexts • The context of a process • Limiting the scope of transitions • Types, permissions, and constraints

Technical requirements Check out the following video to see the Code in Action: https://bit.ly/3m3JzkP

Introduction to SELinux file contexts SELinux file contexts are the most important configuration that a system administrator will have to work with when working with SELinux on the system. Contexts for files are generally identified through a label that is assigned to the file. Mislabeled files are a constant source of headaches for sysadmins, and most common SELinux issues are resolved by correcting the SELinux context. Knowing where and how SELinux contexts are used is key to understanding and resolving SELinux related issues. The following diagram shows how contexts are applied on regular Linux resources, and how the LSM subsystem uses these contexts for decision making:

Introduction to SELinux file contexts

105

Figure 4.1 – Distinction between contexts and regular Linux info

Let's consider a web-based deployment as an example: DokuWiki. This is a popular PHP wiki that uses files rather than a database as its backend system, and is easy to install and manage. As a web hosting platform, we will use nginx.

Getting context information Let's assume that the DokuWiki application will be hosted at /srv/web/localhost/ htdocs/dokuwiki and that it will store its wiki pages (user content) in the data/ subdirectory. We start by downloading the latest DokuWiki tarball from the project site, http://download.dokuwiki.org, and extract it to this location: # mkdir -p /srv/web/localhost/htdocs/ # tar -C /srv/web/localhost/htdocs/ -xvf dokuwiki.tgz # chown -R nginx:nginx /srv/web/localhost/htdocs/dokuwiki

While distributions might have prepackaged DokuWiki installations available, we will use the manual installation approach to show the various file context-related actions in this chapter.

106

Using File Contexts and Process Domains

The contexts of files can easily be acquired using the -Z option of the ls command. Most utilities that can provide feedback on contexts will try to do so using the -Z option, as we saw with the id utility in Chapter 1, Fundamental SELinux Concepts, and Chapter 3, Managing User Logins. Let's look at the current context of the dokuwiki directory itself: # ls -dZ /srv/web/localhost/htdocs/dokuwiki undefined_u:object_r:var_t:s0 /srv/web/localhost/htdocs/ dokuwiki

The context displayed here is var_t. In the Keeping or ignoring contexts section, we will change this to the correct context (as var_t is too generic and not meant for hosting web content). File and directory contexts are stored in the filesystem as extended attributes when the filesystem supports this. An extended attribute (often abbreviated to xattr) is a key/value combination associated with a resource's inode (an information block that represents a file, directory, or symbolic link on a filesystem). Each resource can have multiple extended attributes, but only one value per unique key. When we talk about assigning a label to a file or directory (or relabeling a file), then we imply setting or updating this extended attribute, as it is the label that SELinux will use to obtain the SELinux context for the file. Important note Filesystems that do not support extended attributes can still be used on SELinux-enabled systems. However, the entire filesystem (including all its files and directories) will then be shown with a single context, and differentiation across resources on the filesystem is not possible. We explain how to define file contexts on these filesystems in the Using mount options to set SELinux contexts subsection in this chapter.

By convention, extended attributes on Linux use the following syntax: .=

The namespace of an extended attribute allows for additional access controls or features. Of the currently supported extended attribute namespaces (security, system, trusted, and user), the security namespace enforces specific restrictions on manipulating the attribute: if no security module is loaded (for instance, SELinux is not enabled), then only processes with the CAP_SYS_ADMIN capability (basically root or similarly privileged processes) can modify this parameter.

Introduction to SELinux file contexts

107

We can query the existing extended attributes using the getfattr application, as shown in the following example: $ getfattr -m . -d dokuwiki # file: dokuwiki security.selinux="unconfined_u:object_r:var_t:s0"

As we can see, the security.selinux extended attribute hosts the SELinux context. This ensures that non-administrative users cannot alter the SELinux context of a file when SELinux is disabled and that the SELinux policy controls who can manipulate contexts when SELinux is enabled. The stat application can also be used to show SELinux contexts: $ stat dokuwiki File: dokuwiki Size: 211 Blocks: 0 IO Block: 4096 directory Device: fd01h/64769d Inode: 8512888 Links: 8 Access: (0755/drwxr-xr-x) Uid: ( 0/ root) Gid: ( 0/ root) Context: unconfined_u:object_r:var_t:s0 ...

Getting context information from a file or directory should be as common to an administrator as getting regular access control information (the read (r), write (w), and execute (x) flags).

Interpreting SELinux context types After using SELinux for a while, the motive behind using file labels to assign an SELinux context to the file becomes somewhat clearer. SELinux contexts are named after their purpose, allowing administrators to more easily see whether a context is correctly assigned. Consider the context of a user file in its home directory (user_home_t), a directory in /tmp for a Java application (java_tmp_t), or a socket of rpcbind (rpcbind_ var_run_t). All these files or directories have considerably different purposes on the filesystem, and this reflects itself in the assigned contexts. Policy writers will always try to name the context consistently, making it easier for us to understand the purpose of the file, but also to make the policy almost self-explanatory so that administrators can understand the purpose of the policy without additional documentation needs.

108

Using File Contexts and Process Domains

For the regular filesystem, for instance, files are labeled with a context resembling their main location as they have similar security properties. For example, we find binaries in the /bin folder (and /usr/bin) to be associated with the bin_t type, boot files in /boot associated with boot_t, and generic system resources in /usr associated with usr_t. We can also find more application-specific contexts. For instance, for the PostgreSQL database server, we have the following: • The postgresql_t context is meant for the application itself (process type or domain). • The postgresql_port_t context is meant for the TCP port on which the PostgreSQL daemon listens. • The postgresql_server_packet_t and postgresql_client_packet_t contexts are types associated with network packets received (in case of the postgresql_server_packet_t type) or sent to the PostgreSQL port. • The postgresql_exec_t type is assigned to the postgres binary. • The various postgresql_*_t types for specific filesystem locations related to the daemon, such as postgresql_var_run_t (to apply to resources in /var/run), postgresql_etc_t (to apply to resources in /etc), postgresql_log_t (to apply to resources in /var/log), and postgresql_tmp_t (to apply to resources in /tmp). • The mysqld_db_t type for the database files themselves. Based on the context of a file or resource, administrators can easily detect anomalies in the system setup. An example of an anomaly is when we move a file from the user's home directory to a web server location. When this occurs, the file retains the user_home_t context as extended attributes are moved with it. As the web server process isn't allowed to access user_home_t by default, it will not be able to serve this file to its users. Let's see how to properly set contexts during such copy or move operations.

Keeping or ignoring contexts Now that we are aware that file contexts are stored as extended attributes, how do we ensure that files receive the correct label when they are written or modified? To set an SELinux context on a filesystem resource, a few guidelines exist, ranging from inheritance rules to explicit commands.

Keeping or ignoring contexts

109

Inheriting the default contexts By default, the SELinux security subsystem uses context inheritance to identify which context should be assigned to a file (or directory, socket, and so on) when it is created. A file created in a directory with a var_t context will be assigned the var_t context as well. This means that the file inherits the context from the parent directory and not from the context of the executing process. There are a few exceptions to this though: • SELinux-aware applications can force the context of a file to be different (assuming the SELinux policy allows it, of course). As this is within the software code itself, this behavior cannot be generally configured. • An application called restorecond can be used that enforces contexts on various paths/files based on SELinux's context rules. We will cover these rules and the restorecond application in the SELinux file context expressions and Modifying file contexts sections, respectively. • The SELinux policy allows for transition rules that consider the context of the process creating new files or directories, as well as the name of the file the process is creating. It is these transition rules we will cover next.

Querying transition rules Type transition rules are policy rules that force the use of a different type upon certain conditions. For file contexts, such a type transition rule can be as follows: if a process running in the httpd_t domain creates a file in a directory labeled with the var_log_t SELinux type, then the type identifier of the file becomes httpd_log_t. Basically, this rule assigns the httpd_log_t web server log context to any file placed in a log directory by web servers, rather than the default var_log_t, which would be the case when standard inheritance was used. We can query these type transition rules using sesearch. The sesearch application is one of the most important tools available to query the current SELinux policy. For the previous example, we need the (source) domain and the (target) context of the directory: httpd_t and var_log_t. In the following example, we use sesearch to find the type transition declaration related to the httpd_t domain toward the var_log_t context: $ sesearch -T -s httpd_t -t var_log_t type_transition httpd_t var_log_t:file httpd_log_t;

110

Using File Contexts and Process Domains

The type_transition line is an SELinux policy rule, which maps perfectly to the description. Let's look at another set of type transition rules for the tmp_t type (assigned to the directory used for temporary files, such as /tmp and /var/tmp): $ sesearch -T -s httpd_t -t tmp_t type_transition httpd_t tmp_t:dir httpd_tmp_t; type_transition httpd_t tmp_t:file httpd_tmp_t; type_transition httpd_t tmp_t:file krb5_host_rcache_t HTTP_23; type_transition httpd_t tmp_t:file krb5_host_rcache_t HTTP_48; type_transition httpd_t tmp_t:lnk_file httpd_tmp_t; type_transition httpd_t tmp_t:sock_file httpd_tmp_t;

The policy tells us that, if a file, directory, symbolic link, or socket is created in a directory labeled tmp_t, then this newly created resource gets the httpd_tmp_t context assigned (and thus not the default, inherited tmp_t one). Alongside these rules, it also contains two named file transitions, which are more flexible transition rules. With named file transitions, the policy can consider the name of the file (or directory) created to select a more appropriate context. In the previous example, if a file named HTTP_23 or HTTP_48 is created in a directory labeled tmp_t, then it does not get the httpd_tmp_t context assigned (as would be implied by the regular type transition rules), but the krb5_host_rcache_t type (used for Kerberos implementations) instead. Type transitions not only give us insight into what labels (and thus also SELinux contexts) are going to be assigned, but also give us some clues as to which types are related to a particular domain. In the web server example, we found out by querying the policy that its log files are most likely labeled httpd_log_t, and its temporary files httpd_tmp_t.

Copying and moving files File contexts can also be transferred together with the file itself during copy or move operations. By default, Linux will do the following: • Retain the file context in case of a move (mv) operation on the same filesystem (as this operation does not touch extended attributes, but merely adjusts the metadata of the file). • Ignore the current file context in case of a move operation across a filesystem boundary, as this creates a new file, including content and extended attributes. Instead, it uses the inheritance (or file transition rules) to define the target context. • Ignore the file context in case of a copy (cp) operation, instead using the inheritance (or file transition rules) to define the target context.

Keeping or ignoring contexts

111

Luckily, this is just default behavior (based on the extended attribute support of these utilities) that can be manipulated freely. We can use the -Z option to tell mv that the context of the file should be set to the default type associated with the target location. For instance, in the next example, two files are moved from a user's home directory to the /srv directory. The first example will retain its file context (user_home_t or admin_home_t), while the second one will receive the type associated with user files placed in /srv (var_t): # touch test1 test2 # mv test1 /srv # mv -Z test2 /srv # ls -Z /srv/test* staff_u:object_r:admin_home_t:s0 /srv/test1 staff_u:object_r:var_t:s0 /srv/test2

Similarly, we can tell the cp command through the --preserve=context option to preserve the SELinux context while copying files. Using the same example, we now get the following: # cp test1 /srv # cp --preserve=context test2 /srv # ls -Z /srv/test* staff_u:object_r:var_t:s0 /srv/test1 staff_u:object_r:admin_home_t:s0 /srv/test2

Most of the utilities provided through the coreutils package support the -Z option: mkdir (to create a directory), mknod (to create a device file), mkfifo (to create a named pipe), and so on. Important note If the mv command returns failed to set the security context when using the -Z option, then it is very likely that the location either does not have a valid context associated with it, or that the filesystem does not support SELinux labels. The former is for instances applicable when moving files to /tmp as the CentOS SELinux policy does not have any default context set for files and directories inside /tmp. Newly created resources always need to have their own affiliated labels applied (such as user_ tmp_t).

112

Using File Contexts and Process Domains

Even more so, many of these utilities allow the user to explicitly provide a context through the --context option. For instance, to create a directory, /srv/foo, with the context user_home_t, using mkdir by default would not work, as the target context would be set to var_t. With the --context option, we can tell the utility to set a specific context: # mkdir --context=user_u:object_r:user_home_t:s0 /srv/foo # ls -dZ /srv/foo user_u:object_r:user_home_t:s0 /srv/foo

For other utilities, it is best to consult the manual page and see how the utility deals with extended attributes. For instance, the rsync command can preserve the extended attributes by using the -X or --xattrs option.

Temporarily changing file contexts We can use the chcon tool to update the context of the file (or files) directly. In our previous example, we noticed the var_t label on the DokuWiki files. This is a generic type for variable data and is not the right context for web content. We can use chcon to put the httpd_sys_content_t label on these files, which would allow web servers to have read access on these resources: # chcon -R -t httpd_sys_content_t /srv/web

Another feature that chcon offers is to tell it to label a file or location with the same context as a different file. In the next example, we use chcon to label /srv/web and its resources with the same context as used for the /var/www directory: $ chcon -R --reference /var/www /srv/www

If we change the context of a file through chcon and set it to a context different from the one in the context list, then the context might be reverted later: package managers might reset the file contexts back to their intended value, or the system administrator might trigger a fill filesystem relabeling operation. Until now, we've only focused on the type part of a context. Contexts, however, also include a role part and an SELinux user part. If UBAC is not enabled, then the SELinux user has no influence on any decisions, and resetting it has little value. If UBAC is enabled, though, it might be necessary to reset the SELinux user values on files. Utilities such as chcon can set the SELinux user as well: # chcon -u system_u -R /srv/web

Keeping or ignoring contexts

113

The role for a file is usually object_r as roles currently only make sense for users (processes). To be able to change contexts, we do need the proper SELinux privileges, named relabelfrom and relabelto. These rights are granted on domains to indicate whether the domain can change a label from one type to another. If we find denials in the audit log related to these permissions, then this means that the policy prohibits the domain from changing the contexts.

Placing categories on files and directories We focused primarily on changing types and briefly touched SELinux users, but another important part is to support categories and sensitivity levels. With chcon, we can add sensitivity levels and categories as follows: # chcon -l s0:c0,c2 doku.php

Another tool that can be used to assign categories is the chcat tool. With chcat, we can assign additional categories rather than having to reiterate them, as would be the case with chcon, and even enjoy the human-readable category levels provided by the setrans. conf file: # chcat -- +Contracts doku.php

To remove a category, just use the minus sign: # chcat -- -Contracts doku.php

To remove all categories, use the -d option: # chcat -d doku.php

Users and administrators should keep in mind that applications generally do not set categories themselves, so they need to be added ad hoc.

Using multilevel security on files When the system uses an MLS policy, the chcon tool needs to be used. The syntax is the same as with categories. For instance, to set the sensitivity s1 and category set c2 and c4 to c10 on all files of a user's home directory, you'd do the following: $ chcon -R -l s1:c2,c4.c10 /home/lisa

114

Using File Contexts and Process Domains

Remember that both the context of the user executing chcon and the context of the user who will use the data must be able to deal with the mentioned sensitivity.

Backing up and restoring extended attributes As with the regular file operation tools (such as mv and cp), backup software, too, needs to consider SELinux contexts. Two important requirements exist for a backup tool when working with SELinux-enabled systems: • The backup tool must run in an SELinux context capable of reading all files in scope of the backup, and, of course, of restoring those files as well. If no specific SELinux policy for the backup tool exists, then it might need to run in an unconfined or highly privileged domain to succeed. • The backup tool must be able to back up and restore extended attributes. A popular tool for taking backups (or archives) is the tar application, which supports SELinux contexts as follows: # tar cjvf dokuwiki-20200405.tar.bz2 /srv/web --selinux

When creating a tar archive, add --selinux to include SELinux contexts (both during the creation of the archive and when extracting files from the archive).

Using mount options to set SELinux contexts Not all filesystems support extended attributes. When we use a filesystem without extended attribute support, then the SELinux context of a file is either based on the filesystem type itself (each filesystem has its own associated context) or is passed on to the system using a mount option. The most commonly used mount option in these situations is the context= option. When set, it will use the mentioned context as the context for all the resources in the filesystem. For instance, to mount an external USB drive that hosts a FAT filesystem while ensuring that end users can write to it, we could mount it with the user_home_t context: # mount -o context="user_u:object_r:user_home_t:s0" /dev/sdc1 / media/usb

SELinux file context expressions

115

If the filesystem supports extended attributes but doesn't have all files labeled yet, then we can use the defcontext= option to tell Linux that, if no SELinux context is available, then the default context provided should be used: # mount -o defcontext="system_u:object_r:var_t:s0" /dev/sdc1 / srv/backups

Another mount option is fscontext=. This assigns a context on the filesystem type rather than the context of the files on the filesystem. For instance, a CD/DVD filesystem can be ISO 9660, Joliet, or UDF. SELinux uses this type definition on a filesystem to map permissions such as mount operations and file creation. With the fscontext= option, the filesystem type can be set differently from what the default filesystem type would be. The last option that can be used when mounting filesystems is the rootcontext= option. This will force the root inode of the filesystem to have the given context even before the filesystem is visible to the user space. Permission checks on the location during the mount operation itself can cause havoc when the location does not have the expected context (especially when filesystems are mounted outside their expected location). The rootcontext= option provides a reusable configuration option to set the expected context: # mount -o rootcontext="system_u:object_r:tmp_t:s0" -t tmpfs none /var/tmp

That's it – these are all the context-related mount options. A final note though: the context= option is mutually exclusive to the defcontext= and fscontext= options. So, while the defcontext= and fscontext= options can be used together, they cannot be used with the context= option. Assuming the target filesystem allows for extended attributes, then we can use the file context expressions, which we will cover in the next section.

SELinux file context expressions When we think that the context of a file is wrong, we need to correct the context. SELinux offers several methods to do so, and some distributions even add in more. We can use tools such as chcon, restorecon (together with semanage), setfiles, rlpkg (Gentoo), and fixfiles. Of course, we could also use the setfattr command, but that would be the least user-friendly approach for setting contexts. Let's see how we can set context expressions in a more manageable way.

116

Using File Contexts and Process Domains

Using context expressions In the SELinux policy, a list of regular expressions is kept that informs the SELinux utilities and libraries what the context of a file (or other filesystem resource) should be. Though this expression list is not enforced on the system directly, administrators and SELinux utilities use it to see whether a context is correct, and to reset contexts to what they are supposed to be. You can find the list itself in /etc/selinux/targeted/ contexts/files in the various file_contexts.* files. As an administrator, we can query this list through semanage fcontext as follows: # semanage fcontext -l SELinux fcontext type Context / directory system_u:object_r:root_t:s0 ... /vmlinuz.* symbolic link system_u:object_r:boot_t:s0 /xen(/.*)? all files system_u:object_r:xen_image_t:s0 ...

An example of a tool that queries this information is matchpathcon, which we introduced in Chapter 2, Understanding SELinux Decisions and Logging: # matchpathcon /srv/web/localhost/htdocs/dokuwiki /srv/web/localhost/htdocs/dokuwiki system_u:object_r:var_t:s0

Not all the entries are visible through the semanage application though. Entries related to specific user home directories (such as /home/lisa/.ssh) are not shown as these entries depend on the Linux user (and, more importantly, its associated SELinux user). But for all other entries, the output of the command contains the following: • A regular expression that matches one or more paths • The classes to which the rule is applicable, but translated into a more human-readable format • The context to assign to the resources that match the expression and class list The class list allows us to differentiate contexts based on the resource class. The semanage fcontext output uses human-readable identifiers: resource classes can be a regular file (--), a directory (-d), a socket (-s), a named pipe (-p), a block device (-b), a character device (-c), or a symbolic link (-l). When it says all files, the line is valid regardless of the class.

SELinux file context expressions

117

Right now, we have not defined such rules yet, but after the next section, even defining custom SELinux context expressions will no longer hold any secrets. An important property of the context list is how SELinux prioritizes its application – after all, we could easily have two expressions that both match a certain resource or path. Within SELinux, the most specific rule wins. The logic used is as follows (in order): 1. If line A has a regular expression and line B doesn't, then line B is more specific. 2. If the number of characters before the first regular expression in line A is less than the number of characters before the first regular expression in line B, then line B is more specific. 3. If the number of characters in line A is less than in line B, then line B is more specific. 4. If line A does not map to a specific SELinux type (the policy editor has explicitly told SELinux not to assign a type) and line B does, then line B is more specific. There is a caveat with the rule order, however. When additional rules are added through semanage (which we describe in the next section), then SELinux's utilities apply the rules in the order they were added rather than their specificity. So, instead of the most specific rule, the most recently added rule that matches the path is used.

Registering file context changes Because changing an SELinux context using chcon is often just a temporary measure, it is seriously recommended to only use chcon when testing the impact of a context change. Once the change is acceptable, we need to register it through semanage. For instance, to permanently mark /srv/web (and all its subdirectories) as httpd_sys_ content_t, and the DokuWiki data/ and conf/ folders as httpd_sys_rw_ content_t (to allow the web server to modify these resources), we need to execute the following: # semanage fcontext -a -t httpd_sys_content_t "/srv/web(/.*)?" # semanage fcontext -a -t httpd_sys_rw_content_t "/srv/web/ localhost/htdocs/dokuwiki/data(/.*)?" # semanage fcontext -a -t httpd_sys_rw_content_t "/srv/web/ localhost/htdocs/dokuwiki/conf(/.*)?" # restorecon -Rv /srv/web

118

Using File Contexts and Process Domains

What we do here is register /srv/web and its subdirectories as httpd_sys_ content_t and the two writable directories as httpd_sys_rw_content_t through semanage. Then, we use restorecon to (recursively) reset the contexts of /srv/web to the value registered in the context list. This is the recommended approach for setting contexts on most resources. These registrations are local (custom) context expressions and are stored in a separate configuration file (file_contexts.local). Considering the priority of (locally added) expressions, it is important to have the most specific entries added last, as otherwise the more broadly defined rule for httpd_sys_content_t would be applied to the entire directory. This is unlike the priority rules for (policy added) expressions that do have the concept of most specific rule wins. The semanage fcontext application can also be used to inform SELinux that a part of the filesystem tree should be labeled similarly as a different location on the filesystem. Such an equivalency rule allows us to use different paths for application installations or file destinations and tell semanage to apply the same contexts as if the destination were the default. Let's make this more visible through an example, and have everything under /srv/web be labeled in a similar manner to the files at /var/www (including subdirectories), so /srv/web/icons gets the same context as /var/www/icons. We use the -e option of semanage fcontext to create such an equivalency as follows: # semanage fcontext -a -e /var/www /srv/web # restorecon -Rv /srv/web

This will create a substitution entry so that anything under /srv/web gets the same label as if it were at the same location under /var/www. Most distributions already configure a few equivalency rules that we can read as follows: # cat /etc/selinux/targeted/contexts/files/file_contexts.subs_ dist /run /var/run ... /sysroot/tmp /tmp

The semanage fcontext -l command will show these equivalent locations at the end of its output as well.

SELinux file context expressions

119

Optimizing recursive context operations The restorecon application resets the SELinux context of files and other resources based on the context definitions managed through the SELinux policy and semanage fcontext. When applying restorecon in a recursive fashion against directories, this might take a while. To improve performance in this situation, the SELinux authors support the skipping of restorecon operations. With the -D option to restorecon, an additional extended attribute will be written to the main directory that contains a hash of the file context definitions used when invoking the command: # restorecon -RD /home

Subsequent invocations of restorecon with -D will check this hash to see whether any of the file context definitions that impact this directory have been modified (using semanage fcontext). If there aren't, then the restore operation will be skipped: # restorecon -RvD /home Skipping restorecon as matching digest on: /home

Once we update a definition that influences the given location, then restorecon will reset the contexts appropriately: # semanage fcontext -a -t httpd_user_content_t "/home/[^/]*/ cgi-bin(/.*)?" # restorecon -RvD /home Relabeled /home/lisa/cgi-bin from staff_u:object_r:user_ home_t:s0 to staff_u:object_r:httpd_user_content_t:s0 Updated digest for: /home

The restorecon_xattr command can be used to manage these extended attributes (view or delete) and show how the attributes are formed: # restorecon_xattr -v /home specfiles SHA1 digest: 7ed69be330ad60811481e455ca8e5ab0b1556036 calculated using the following specfile(s): /etc/selinux/targeted/contexts/files/file_contexts.subs_dist ... /etc/selinux/targeted/contexts/files/file_contexts.local.bin /home Digest: 7ed69be330ad60811481e455ca8e5ab0b1556036 Match

120

Using File Contexts and Process Domains

The digest referenced is the security.restorecon_last or security.sehash extended attributes. More recent user space tools use the latter, and apply their logic to each subdirectory, whereas older user space utilities use the former and only apply their logic on the selected directory. The disadvantage of the security.restorecon_last usage is that it does not work with subdirectories: if we apply a recursive restorecon operation against /, then this tool will ignore the digest on /home. With the security.sehash usage, a recursive operation against / will check the digest for /home as well.

Using customizable types Some SELinux types are meant for files whose paths cannot be accurately defined by administrators or where the administrator does not want the context to be reset when a relabeling operation is triggered. For these purposes, SELinux supports what it calls customizable types. When tools that manage file contexts (such as restorecon) encounter a file with a customizable type set, they will not revert its context to the registered context definition. The customizable types are declared in the customizable_types file inside / etc/selinux/targeted/contexts. To have restorecon relabel such files, administrators need to pass the force reset option (-F) before the tool resets the contexts. Let's look at the contents of this customizable_types file: $ cat /etc/selinux/targeted/contexts/customizable_types container_file_t sandbox_file_t ... httpd_user_content_t git_session_content_t home_bin_t user_tty_device_t

As an example, we can mark a file in a home directory as home_bin_t, which is a customizable type, and as such, this file will not be relabeled back to user_home_t when a filesystem relabeling operation is done: $ chcon -t home_bin_t ~/convert.sh

SELinux file context expressions

121

Marking other types as customizable requires updating the customizable_types file, as there is no user command that adds or removes type definitions from this list. Because this file can be overwritten when the distribution or administrator pushes out a new policy package, it needs to be governed carefully. That said, the use of customizable types has its advantages. As an administrator, we might want to create and support specific types as usable by end users who can use chcon to set the contexts of individual files in their home directory. By having those types marked as customizable types, a relabeling operation against /home will not reset those contexts. When the target type is not a customizable type, administrators generally prefer to use semanage fcontext to add an expression and restorecon to fix the context of the files. Most administrators will use directory-based labeling: this is much easier to maintain, and much easier to explain to end users. Many will even use this approach for customizable types: # semanage fcontext -a -t home_bin_t "/home/[^/]*/bin(/.*)?"

With this command, user binaries and scripts located in the ~/bin directory will be labeled as home_bin_t.

Compiling the different file_contexts files Inside the /etc/selinux/targeted/contexts/files directory, five different file_contexts files can be found: • The file_contexts file itself (without any suffix) is the basic expression file provided by the SELinux policy offered through the Linux distribution. • The file_contexts.local file contains the locally added rules (through the semanage fcontext command, which we covered earlier in this chapter). • The file_contexts.homedirs file contains the expressions for the user home directories. When new user mappings are created and managed through semanage login and semanage user, this file is adjusted to reflect the new situation. • The file_contexts.subs_dist file contains equivalency rules, provided by the distribution's SELinux policy, which tell SELinux to consider one part of the filesystem as having the same labeling rules as another location. • The file_contexts.subs file contains locally managed equivalency rules (through the semanage fcontext command, covered earlier in this chapter).

122

Using File Contexts and Process Domains

Alongside those files, you will find associated *.bin files (so file_contexts. bin for the file_contexts file, file_contexts.local.bin for the file_ contexts.local file, and so on). These *.bin files are automatically created, but in case of a discrepancy, administrators can rebuild the files themselves as well using the sefcontext_compile command: # cd /etc/selinux/targeted/contexts/files # sefcontext_compile file_contexts.local

These files contain the same information as the main file, but are precompiled to make lookups faster. Unless the tools detect that the *.bin files are older than their source files, the SELinux utilities will use the compiled versions of these files.

Exchanging local modifications When local modifications are registered through semanage fcontext, they only apply to a single system. If local definitions need to be reapplied on various systems, administrators can extract the local modifications and import them on another system. To export the local modifications, use semanage export: # semanage export -f local-mods.conf

The file that contains the local modifications (local-mods.conf in the example) can be adjusted at will. This allows administrators to remove all lines except those they want to apply on other systems. With the local modifications stored in the file, transport the file to the other system(s) and import the settings: # semanage import -f ./local-mods.conf

The imported settings are immediately registered. Of course, in case of filesystem changes (semanage fcontext), don't forget to run restorecon against the target directories.

Modifying file contexts We now know how to set SELinux contexts, both directly through tools such as chcon as well as through the restorecon application, which queries the SELinux context list to know what context a file should have. Yet restorecon is not the only application that considers this context list.

Modifying file contexts

123

Using setfiles, rlpkg, and fixfiles The setfiles application is an older one, which requires the path to the context list file itself to reset contexts. It is often used under the hood of other applications, so most administrators do not need to call setfiles directly anymore: # setfiles /etc/selinux/targeted/contexts/files/file_contexts / srv/web

Another set of tools are the rlpkg (Gentoo) and fixfiles (CentOS and related distributions) applications. Both these applications have a nice feature: they can be used to reset the contexts of the files of an application rather than having to iterate over the files manually and run restorecon against them. In the next example, we're using these tools to restore the contexts of the files provided by the nginx package: # rlpkg nginx # fixfiles -R nginx restore

Another feature of both applications is that they can be used to relabel the entire filesystem without the need to perform a system reboot, like so: # rlpkg -a -r # fixfiles -f -F relabel

Of course, this is not as fine-grained as the commands before.

Relabeling the entire filesystem The rlpkg and fixfiles commands as listed in the previous section are not the only available approaches for relabeling the entire filesystem when working with a CentOS (or related) distribution. SELinux offers two other methods to ask the system to perform a full filesystem relabeling operation during (re)boot: placing a touch file (which the system reads at boot time) or configuring a boot parameter. The touch file is called .autorelabel and should be placed in the root filesystem. Once set, the system needs to be rebooted: # touch /.autorelabel # reboot

124

Using File Contexts and Process Domains

We trigger the same behavior if we add the autorelabel=1 parameter to the boot parameter list (like where we can set the selinux= and enforcing= parameters as discussed earlier). Asking the system to perform a full filesystem relabeling operation will take a while. When finished, the system will reboot again. Touch files will be removed automatically after the relabeling operation has finished.

Automatically setting context with restorecond Contexts can also be applied by the restorecond daemon. The purpose of this daemon is to enforce the expression list rules onto a configurable set of locations, defined in the /etc/selinux/restorecond.conf file. The following set of files and directories is an example list of locations configured in the restorecond.conf file so that restorecond automatically applies the SELinux contexts on these files and directories whenever it detects a context change in them: /etc/services /etc/resolv.conf /etc/samba/secrets.tdb ... /root/.ssh/*

In this case, if a process creates a file that matches any of the previously created paths, the Linux inotify subsystem will notify restorecond of it. restorecond will then relabel the file according to the expression list, applying the correct label regardless of the process (and context) that created the file. The use of restorecond is primarily for historical reasons, when SELinux didn't support named file transitions. At that time, writing resolv.conf in /etc could not be differentiated from writing to the passwd file in /etc. The introduction of named file transitions has considerably reduced the need for restorecond.

Setting SELinux context at boot with tmpfiles If the Linux distribution uses systemd, then you can use systemd-tmpfiles to automatically set SELinux context at boot. systemd uses the tmpfiles application to automatically create and manage volatile locations on the system, such as locations inside /run when /run is a tmpfs-mounted filesystem (an in-memory filesystem).

The context of a process

125

Administrators can configure tmpfiles to automatically create files, directories, device files, symbolic links, and others at boot, and to reset the permissions on resources. It is through this reset operation that we can use tmpfiles to set the right SELinux context at boot time. In Chapter 3, Managing User Logins, we covered polyinstantiation, where users get their own private view on filesystem resources. The example we gave used a directory called /tmp/tmp-inst, which had to have the 000 permission set, and which will host the user-oriented /tmp views. Rather than having to create and set this permission each time, we can configure tmpfiles to do this for us, and define the right SELinux context up front: # semanage fcontext -a -t tmp_t -f d "/tmp/tmp-inst"

In /etc/tmpfiles.d, we create a file called selinux-polyinstantiation.conf with the following content: d /tmp/tmp-inst 000 root root

The name of the file can be chosen freely, but make sure it uses the .conf suffix. Every time the system boots, systemd-tmpfiles will ensure that the /tmp/tmp-inst directory is created with the appropriate permissions. If a location does not need to be created, but only its SELinux context reset, then you can use the z (one resource) or Z (recursively) options in the tmpfiles configuration. This is used, for instance, by the default SELinux tmpfiles configuration, selinux-policy. conf, in /usr/lib/tmpfiles.d: z /sys/devices/system/cpu/online - - -

The - used is to inform tmpfiles not to adjust the permissions and ownership, and only to reset the SELinux context.

The context of a process As everything in SELinux works with contexts, even processes are assigned a context, also known as the domain. Let's see how we can obtain this information, how SELinux transitions from one domain to another, and learn how to query the SELinux policy to find more information about these transitions.

126

Using File Contexts and Process Domains

Getting a process context We saw that the nginx web server runs in the httpd_t domain, which can be seen with the ps -eZ command, as follows: # ps -eZ | grep nginx system_u:system_r:httpd_t:s0

3744 ?

00:00:00 nginx

Several other ways exist to obtain the process context. Although the method with ps is the most obvious, these other methods can prove useful in scripted approaches or through monitoring services. A first approach is to read the /proc//attr/current pseudo-file, which we've already encountered in Chapter 1, Fundamental SELinux Concepts. It displays a process's current security context: # pidof nginx 3746 3745 3744 # cat /proc/3744/attr/current system_u:system_r:httpd_t:s0

To receive a somewhat more human-readable output, use the secon command for the given process ID: # secon --pid 3744 user: system_u role: system_r type: httpd_t sensitivity: s0 clearance: s0 mls-range: s0

Finally, the SELinux user space project has a helper utility called getpidcon, which the libselinux library optionally provides. Although this utility is not available on CentOS (or related distributions), other distributions such as Gentoo do have it. The utility requires a single PID and returns its context: # getpidcon 679 system_u:system_r:nginx_t:s0

Now, the Apache processes don't themselves inform SELinux that they need to run in the httpd_t (or, for Gentoo, the nginx_t) domain. For that, transition rules exist in the SELinux policy that govern when and how processes are executed in a specific domain.

The context of a process

127

Transitioning toward a domain Just as we have seen with files, if a process forks and creates a new process, this process, by default, inherits the context of the parent process. For the web server, the main process is running in the httpd_t domain, so all the launched worker processes inherit the httpd_t domain from it. To differentiate the domain of one process from another, domain transitions can be defined. A domain transition (also known as a process transition or type transition) is a rule in SELinux that tells SELinux another domain is to be used for a forked process (actually, it is when the parent process calls the execve() function, most likely after a fork() operation). Like the file-based transitions, domain transitions can be queried using sesearch. Let's investigate the domains allowed to transition to the httpd_t domain: $ sesearch -T -t httpd_exec_t type_transition certwatch_t httpd_exec_t:process httpd_t; type_transition cluster_t httpd_exec_t:process httpd_t; type_transition initrc_t httpd_exec_t:process httpd_t; ... type_transition system_cronjob_t httpd_exec_t:process httpd_t;

In this case, SELinux will switch the context of a launched web server to httpd_t if the parent process is running in one of the mentioned domains (such as the initrc_t domain) and is executing a file labeled as httpd_exec_t (the label assigned to the httpd and nginx binaries). But for this to truly happen, several other permissions (next to the domain transition) need to be in place. The following list describes these various permissions: • The source process (such as initrc_t) needs to be allowed to transition to the httpd_t domain, governed by the transition privilege on the process class: $ sesearch -s initrc_t -t httpd_t -c process -p transition -A

• The source process (such as initrc_t) needs to have the right of execution on the file it is launching (httpd_exec_t): $ sesearch -s initrc_t -t httpd_exec_t -c file -p execute -A

128

Using File Contexts and Process Domains

• The httpd_exec_t type must be identified as an entry point for the httpd_t domain. SELinux uses an entry point to ensure that a domain transition only occurs when using the specified file context on the executing binary or script: $ sesearch -s httpd_t -t httpd_exec_t -c file -p entrypoint -A

• The target domain must be allowed for the role that the parent process is in. In the case of system daemons, the role is system_r: $ seinfo -r system_r -x | grep httpd_t

A graphical representation of these rights is as follows:

Figure 4.2 – Graphical overview of the necessary transition permissions

Only when all these privileges are allowed will a domain transition occur. If not, then either the execution of the application fails (if the domain has no execute or execute_ no_trans rights on the file), or it executes but remains running in the same domain as the parent process.

The context of a process

129

Domain transitions are an important concept as they inform the administrator how an application gets into its privileged context. To analyze this, many security administrators look at how one context can transition to another. We explain policy analysis in Chapter 13, Analyzing Policy Behavior. For policy writers, deciding when to create a domain transition and when to keep the processes running in the same (source) context is a matter of design. Generally, policy developers will try to keep the parent context confined so that every additional privilege is a source of consideration for switching to another domain (which has that privilege). Basically, policy developers will trigger a transition when the target application requires significantly more (or different) permissions than the source domain holds. That is also why the unconfined_t domain has fewer transitions when executing user applications compared to the confined user domains, user_t or guest_t: the unconfined_t domain already holds many privileges, so transitioning to a different domain has little value. Note that this is a decision made by the policy writers or Linux distribution, not by the SELinux technology itself. All SELinux does is enforce the policy rules.

Verifying a target context When executing applications, the SELinux policy might have the command run in a different domain. Although we could start querying all rules with sesearch, a simpler command exists that tells us what the target context is when we execute a command or script: selinuxexeccon. This command requires at least one argument (the path of the binary or script that would be executed) and an optional second (the source context). If we omit the second argument, the tool will use the current context as the source context. For instance, to find out in which domain the passwd command would run when executed from the current context, we'd use this command: # selinuxexeccon /usr/bin/passwd unconfined_u:unconfined_r:passwd_t:s0-s0:c0.c1023

The following example shows the target context when the init_t domain executes the nginx binary: # selinuxexeccon /usr/sbin/nginx system_u:system_r:init_t:s0 system_u:system_r:httpd_t:s0

Using selinuxexeccon is much faster than querying all appropriate permissions separately.

130

Using File Contexts and Process Domains

Other supported transitions Regular domain transitions are the most common transitions in SELinux, but other transitions are possible as well. For instance, some applications (such as cron or login) are SELinux-aware and will specify which domain to transition to. These applications call the setexeccon() method (set execution context) to specify the target domain and do not use a type transition rule. The other privilege requirements, however, still hold. Some SELinux-aware applications are even able to change their current context (and not just the context of the application they execute). To accomplish this, the application domain needs the dyntransition privilege (one of the privileges supported for process-level activities). One example of such an application is OpenSSH, which, by default, runs in the sshd_t domain but can transition to the sftpd_t type.

Querying initial contexts When SELinux does not have a label yet for a resource, it will assign an initial context (or initial security ID (SID)) to the resource. For a few classes, the SELinux policy will have a default initial context from which it can further jumpstart and assign labels. The initial contexts for various SIDs can be queried using seinfo: # seinfo --initalsid -x Initial SIDs: 27 sid any_socket system_u:object_r:unlabeled_t:s0 sid devnull system_u:object_r:null_device_t:s0 ... sid unlabeled system_u:object_r:unlabeled_t:s0

As you can see, not all classes have a default context assigned, as other classes have their contexts derived from the contexts of the currently listed initial SIDs.

Tweaking memory protections Legacy binaries on Linux systems might require execution permissions to be set on memory regions when these are used for reading, even when the execute permission is not actually used. This read-implies-exec is a nuisance for mandatory access controls such as SELinux because they need to document the appropriate permissions in their policy. If an application needs read access, does the policy then also have to include the implied execute rights? And if the policy does not include execute rights, should the read operation then fail because it implied execute permissions?

Limiting the scope of transitions

131

Informational note Read-implies-exec is a legacy support for running old binaries or binaries compiled for other Unix systems where applications do not explicitly mark their executable memory as executable, assuming that every memory region that is marked as readable is executable. This creates a security risk as malicious actors can load in executable code dynamically without the system being able to prevent the application to execute this code. Many operating systems nowadays have clear memory protection routines in place, including preventing data from becoming executable. Sadly, we often need to deal with legacy situations, so all operating systems have methods in place that selectively disable these memory controls, and within Linux this is done through its personalities support (see man personality for more information).

SELinux developers allow administrators to select their most appropriate permission handling by introducing a memory protection check that can be tuned. The checkreqprot option can be set to 0 to check protections as handled by the kernel, or 1 to check protections as asked by the application. On older systems, this option will be set to 1 to support these legacy binaries. Recent distributions, however, build their applications appropriately, and the more secure setting 0 is used, as displayed by the sestatus command: # sestatus | grep Memory Memory protection checking:

actual (secure)

You can toggle this support through /sys/fs/selinux/checkreqprot: # echo 1 > /sys/fs/selinux/checkreqprot # sestatus | grep Memory Memory protection checking: requested (insecure)

The parameter's default value is configured when building the Linux kernel, through the CONFIG_SECURITY_SELINUX_CHECKREQPROT_VALUE kernel configuration parameter. Administrators can also boot the system with the checkreqprot= boot parameter to have the specified value set.

Limiting the scope of transitions For security reasons, Linux systems can reduce the ability of processes to gain elevated privileges under certain situations or provide additional constraints to reduce the likelihood of vulnerabilities to be exploitable. SELinux developers, too, honor these situations.

132

Using File Contexts and Process Domains

Sanitizing environments on transition When we execute a higher-privileged command (be it a setuid application or one where capabilities are added to the session), the GNU C library (glibc) will sanitize the environment. This means that a set of security-sensitive environment variables are discarded to make sure that attackers, malicious persons, or malicious applications cannot negatively influence the session. This secure execution is controlled through an Executable and Linkable Format (ELF) auxiliary vector called AT_SECURE. When set, environment variables such as LD_PRELOAD, LD_AUDIT, LD_DEBUG, TMPDIR, and NLSPATH are removed from the session. SELinux will force this sanitation on domain transitions as well, ensuring that the newly executed domain does not have access to these sensitive environment variables. Of course, sometimes the transitioned domain requires these variables. Not all domains can deal with sanitized environments, or use these environment variables to pass along important information, so always dropping the environment variables might result in unusable application domains. To allow transitions without sanitizing the environment, the noatsecure permission can be granted to domain transitions. For instance, let's consider the execution of a Firefox plugin: # sesearch -t mozilla_plugin_t -p noatsecure -A ... allow unconfined_t mozilla_plugin_t:process { ... noatsecure ...}; ...

When an application running in the unconfined_t domain executes the plugin (which results in a domain transition to mozilla_plugin_t), the environment variables need to be kept as otherwise the plugin might not function properly. As such, the SELinux policy grants the noatsecure permission to the domains that invoke Firefox plugins.

Disabling unconstrained transitions A second security constraint that Linux supports is to mount a filesystem with the nosuid option. When set, no setuid and setgid binaries on that filesystem will have any effect on the effective user or group ID of the executing session. Essentially, a setuid application on a filesystem mounted with nosuid will act as if no setuid bit is set.

Limiting the scope of transitions

133

To ensure that transitions triggered by applications hosted on a nosuid-mounted filesystem do not allow for elevated privileges, SELinux policy developers must explicitly mark a transition as allowed for nosuid-mounted filesystems, using the nosuid_transition permission. This permission is part of the process2 class: $ sesearch -s unconfined_t -p nosuid_transition -A allow unconfined_t initrc_t:process2 { nnp_transition nosuid_ transition }; ...

This allows policy developers to differentiate regular domain transitions from nosuid-constrained domain transitions. Note SELinux has a limit on the number of privileges that can be assigned to a class. When the number of privileges exceeds 32, the SELinux developers will create a different class and the permissions continue in this second class. Right now, the two classes that have more than 32 permissions are the capability class and the process class.

This permission-based approach might not be in place on all SELinux-enabled systems though. It is enabled when the nnp_nosuid_transition policy capability is defined and set to 1: # cat /sys/fs/selinux/policy_capabilities/nnp_nosuid_transition 1

If this capability value is 0, then SELinux will use a concept called type bounds to support domain transitions for applications hosted on nosuid-mounted filesystems. Any executable with a file context that would result in a domain transition will only result in a domain transition if the target domain is bounded by the parent domain. Note Policy capabilities cannot be tweaked by administrators. They are used by policy developers to inform the Linux kernel which behavior it expects. For the nnp_nosuid_transition capability, the policy developer informs the kernel that the nosuid_transition and nnp_transition permission checks should be used rather than bounded domains, and that its policy will generally only include support for the transitions and not for the bounded domains.

134

Using File Contexts and Process Domains

If it is not bounded, then the domain transition will not occur, and the session will remain in the current context (or the command will fail to execute if the application is not allowed to run in the current context). A bounded domain is not just calculated live based on the permissions though. SELinux has an explicit rule that enforces a target domain to be bounded by a parent domain. Even when permissions are later added to the bounded domain, they will be denied by the SELinux security subsystem if they aren't part of the parent domain. With seinfo, these type bounds can be listed as follows: # seinfo --typebounds

Most distributions, however, do not have bounded domains defined in their SELinux policy anymore, as the new nosuid_transition permission is much more flexible. The use of bounded domains required policy developers to extend the permissions of the parent domain every time the child domain needed to be extended, which was a major nuisance when the parent domain is a generic one (be it a container management platform or a system service daemon).

Using Linux's NO_NEW_PRIVS The use of filesystems mounted with nosuid is a specific case of Linux's No New Privilege (NNP) support. NNP is a process-specific attribute that tells the Linux kernel that the process is no longer to be granted additional privileges. From that point onward, the constraints as mentioned before hold, and SELinux will only allow domain transitions if it has the nnp_transition permission, or toward a bounded domain if the nnp_nosuid_transition policy capability is not set. The parameter can be set by applications themselves using the process control function prctl(), but the user can also influence this. The setpriv command can be used to launch applications with PR_SET_NO_NEW_PRIVS set (the parameter that applications can pass through the prctl() function). As an example, create the following simple Python-based CGI script in a cgi-bin directory inside a regular user's home directory: #!/usr/bin/env python3 import sys, time import subprocess import cgi, cgitb cgitb.enable() print('Content-Type: text/html;charset=utf-8\n') PIPE = subprocess.PIPE

Limiting the scope of transitions

135

STDOUT = subprocess.STDOUT pd = subprocess.Popen(['ping', '-c', '1', 'localhost'], stdout=PIPE, stderr=STDOUT) while True: output = pd.stdout.read(1) if output == '' and pd.poll() != None: break if output != '': sys.stdout.write(output.decode('utf-8')) sys.stdout.flush()

With this CGI script now available, first launch a simple CGI-capable web server (we will pick port 6020 as unprivileged users should be able to bind processes to this port) and connect to it: $ python3 -m http.server --cgi 6020

In a different session, connect to the web server and call the newly created Python script (here named test.py): $ curl http://localhost:6020/cgi-bin/test.py PING localhost(localhost(::1)) 56 data bytes ...

Now, launch the same CGI-capable web server, but with NNP enabled: $ setpriv --no-new-privs python3 -m http.server --cgi 6020

Again, connect to the web server and call the test.py CGI script: $ curl http://localhost:6020/cgi-bin/test.py ping: socket: Permission denied

Because Linux's NNP is enabled, the ping command is not able to obtain the higher privileges needed to open the socket. Sometimes, you'll notice a denial for the execute_no_trans permission in the SELinux audit logs. This occurs when the SELinux policy does not allow an application to be executed without transitioning.

136

Using File Contexts and Process Domains

Types, permissions, and constraints Now that we know more about types (for processes, files, and other resources), let's explore how these are used in the SELinux policy in more detail.

Understanding type attributes We have discussed the sesearch application already and how it can be used to query the current SELinux policy. Let's look at a specific process transition: $ sesearch -s initrc_t -t httpd_t -c process -p transition -A allow initrc_domain daemon:process transition;

Even though we asked for the rules related to the initrc_t source domain and the httpd_t target, we get a rule back for the initrc_domain source domain and the daemon target. What sesearch did here was show us how the SELinux policy allows the requested permission, but through attributes assigned to the initrc_t and httpd_t types. Type attributes in SELinux are used to group multiple types and assign privileges to those groups rather than having to assign the privileges to each type individually. For initrc_domain, the following types are all tagged with this attribute, as can be seen through the seinfo application: $ seinfo -a initrc_domain -x Type Attributes: 1 attribute initrc_domain; cluster_t; ... initrc_t; ... piranha_pulse_t;

As we can see, the initrc_t type is indeed one of the types tagged with initrc_domain. Similarly, the daemon attribute is assigned to several types (several hundred, even). So, the single allow rule mentioned earlier consolidates more than a thousand rules into one. Attributes are increasingly used in the policy as a way of consolidating and simplifying policy development. With seinfo -a, you can get an overview of all the attributes supported in the current policy.

Types, permissions, and constraints

137

Querying domain permissions The most common rules in SELinux are the allow rules, informing the SELinux subsystem what permissions a domain has. allow rules use the following syntax: allow : ;

The field is almost always a domain, whereas the field can be any type. The field allows us to differentiate privileges based on the resource, whether it is for a regular file, a directory, a TCP socket, a capability, and so on. A full overview of all supported classes can be obtained from seinfo -c. Each class has a set of permissions assigned to it that SELinux can control. For instance, the sem class (used for semaphore access) has the following permissions associated with it: $ seinfo -c sem -x Classes: 1 class sem inherits ipc

The reference to ipc in the output informs us that the class inherits permission from the common ipc class, which we can query as follows: $ seinfo --common=ipc -x Commons: 1 { write destroy ... create }

In the field, most rules will bundle a set of permissions using curly brackets: allow user_t etc_t : file { ioctl read getattr lock execute execute_no_trans open };

138

Using File Contexts and Process Domains

This syntax allows policy developers to make very fine-grained permission controls. We can use the sesearch command to query these rules. The more options are given to the sesearch command, the finer-grained our search parameters become. For instance, sesearch -A would give us all allow rules currently in place. Adding a source (-s) filters the output to only show the allow rules for this domain. Adding a destination or target (-t) filters the output even more. Other options that can be used to filter through allow rules with sesearch are the class (-c) and permission (-p) options. As you might have guessed by now, sesearch is an extremely versatile command for querying the active policy, showing us the SELinux policy rules that match the options given.

Learning about constraints The allow statements in SELinux, however, only focus on type-related permissions. Sometimes though, we need to restrict certain actions based on the user or role information. SELinux supports this through constraints. Constraints in SELinux are rules applied against a class and a set of its permissions that must be true for SELinux to further allow the request. Consider the following constraint on process transitions: constrain process { transition dyntransition noatsecure siginh rlimitinh } ( u1 == u2 or ( t1 == can_change_process_identity and t2 == process_user_target ) or ( t1 == cron_source_domain and ( t2 == cron_job_domain or u2 == system_u ) ) or ( t1 == can_system_change and u2 == system_u ) or ( t1 == process_uncond_exempt ) );

Summary

139

This constraint says that at least one of the following rules must be true if a transition, dyntransition, or any of the other three mentioned process permissions is invoked: • The SELinux user of the source (u1) and that of the target (u2) must be the same. • The SELinux type of the source (t1) must have the can_change_process_ identity attribute set, and the SELinux type of the target (t2) must have the process_user_target attribute set. • The SELinux type of the source (t1) must have the cron_source_domain attribute set, and either the target type (t2) should have cron_job_domain as an attribute, or the target SELinux user (u2) should be system_u. • The SELinux type of the source (t1) must have the can_system_change attribute set, and the SELinux user of the target (u2) must be system_u. • The SELinux type of the source (t1) must have the process_uncond_exempt attribute set. It is through constraints that UBAC is implemented as follows: u1 or or or or

== u1 u2 t1 t2

u2 == == != !=

system_u system_u ubac_constrained_type ubac_constrained_type

You can list the currently enabled constraints using seinfo --constrain. Multiple constraints can be active for the same class and permission set. In that case, all the constraints need to be true for the permission to go through.

Summary In this chapter, we learned how file contexts are stored as extended attributes on the filesystem and how we can manipulate the contexts of files and other filesystem resources. Next, we found out where SELinux keeps the definitions that describe which SELinux contexts to assign to the files. We also learned to work with the semanage tool to manipulate this information and worked with a few tools that use this information to enforce contexts on resources.

140

Using File Contexts and Process Domains

On the process level, we got our first taste of SELinux policies, identifying when a process launches inside a certain SELinux domain. With it, we covered the sesearch and seinfo applications to query the SELinux policy. Finally, we looked at some of Linux's security implementations that limit the transition scope of applications, which also influences SELinux domain transitions. In the next chapter, we will expand our knowledge of protecting the operating system through the networking-related features of SELinux.

Questions 1. What is the most common option for Linux tools to display or explicitly set SELinux contexts? 2. How is an SELinux context for a file or directory stored on the system? 3. Why is chcon not recommended to persist SELinux context changes? 4. Is the order of context definitions using the semanage fcontext command important? 5. How do you relabel files on the filesystem? 6. What privileges does a domain need before it can transition to another domain? 7. How do SELinux policies bundle multiple types together to facilitate policy development?

5

Controlling Network Communications The SELinux mandatory access controls go much beyond its file and process access controls. One of the features provided by SELinux is its ability to control network communications. By default, general network access controls use the socket-based access control mechanism, but more detailed approaches are also possible. In this chapter, we will learn how network access controls are governed by SELinux, cover what administrators can do to further strengthen network communications using iptables, and describe how SELinux policies can be used for cross-system security through labeled IPsec. We'll finish the chapter with an introduction to CIPSO labeling and its integration with SELinux. We cover the following topics in this chapter: • Controlling process communications • Linux firewalling and SECMARK support • Securing high-speed InfiniBand networks • Understanding labeled networking • Using labeled IPsec with SELinux • Supporting CIPSO with NetLabel and SELinux

142

Controlling Network Communications

Technical requirements Not all sections in this chapter apply to all environments. For InfiniBand support, for instance, InfiniBand hardware is needed, whereas for NetLabel/CIPSO support, the network in its entirety needs to support the CIPSO (or CALIPSO in the case of IPv6) protocol for the hosts to be able to communicate with each other. Check out the following video to see the Code in Action: https://bit.ly/34bVDdm

Controlling process communications Linux applications communicate with each other either directly or over a network. But the difference between direct communication and networked communication, from an application programmer's point of view, is not always that big. Let's look at the various communication methods that Linux supports and how SELinux aligns with them.

Using shared memory The least network-like method is the use of shared memory. Applications can share certain parts of the memory with each other and use those shared segments to communicate between two (or more) processes. To govern access to the shared memory, application programmers can use mutual exclusions (mutexes) or semaphores. A semaphore is an atomically incremented or decremented integer (ensuring that two applications do not overwrite each other's values without knowing about the value change), whereas a mutex can be interpreted as a special semaphore that only takes the values 0 or 1. On Linux, two implementations exist for shared memory access and control: SysV-style and POSIX-style. We will not dwell on the advantages and disadvantages of each, but rather look at how SELinux governs access to these implementations. SELinux controls the SysV-style primitives through specific classes: sem for semaphores and shm for shared memory. The semaphores, mutexes, and shared memory segments inherit the context of the first process that creates them. Administrators who want to control the SysV-style primitives can use the various ipc* commands: ipcs (to list), ipcrm (to remove), and ipcmk (to create).

Controlling process communications

143

For instance, let's first list the resources and then remove the listed shared memory: # ipcs ... ------ Shared Memory Segments -----key shmid owner perms bytes nattch 0x0052e2c1 0 postgres 600 56 6 # ipcrm -m 0

status

When POSIX-style semaphores, mutexes, and shared memory segments are used, SELinux controls those operations through the file-based access controls. The POSIX-style approach uses regular files in /dev/shm, which is simpler for administrators to control and manage.

Communicating locally through pipes A second large family of communication methods in operating systems is the use of pipes. As the name implies, pipes are generally one-way communication tunnels, with information flowing from one (or more) senders to one receiver (there are exceptions to this, such as Solaris pipes, which act as bidirectional channels, but those are not supported on Linux). Another name for a pipe is first-in, first-out (FIFO). We have two types of pipes in Linux: anonymous pipes (also known as unnamed pipes) and named pipes. The difference is that a named pipe uses a file in the regular filesystem as its identification, whereas anonymous pipes are constructed through the applications with no representation in the regular filesystem. In both cases, SELinux will see the pipes as files of the fifo_file class. Named pipes will have their path associated with the regular filesystem and are created using the mknod or mkfifo commands (or through the mkfifo() function when handled within applications). Anonymous pipes, however, will be shown as part of the pipefs filesystem. This is a pseudo filesystem, not accessible to users, but still represented as a filesystem through Linux's virtual file system (VFS) abstraction. From an SELinux policy point of view, the FIFO file is the target for which the access controls apply: two domains that both have the correct set of privileges toward the context of the FIFO file will be able to communicate with each other.

144

Controlling Network Communications

Administrators can find out which process is communicating over FIFOs with other processes through tools such as lsof, or by querying the /proc filesystem (as part of the /proc//fd listings). The lsof tool supports the -Z option to show the SELinux context of the process, and even supports wildcards: # lsof -Z *:postfix_*

In this example, lsof displays information about all processes that use a postfix_* label.

Conversing over UNIX domain sockets With pipes supporting one-way communication only, any conversation between two processes would require two pipes. Also, true client/server-like communication with pipes is challenging to implement. To accomplish the more advanced communication flows, processes will use sockets. Most administrators are aware that TCP and UDP communication occurs over sockets. Applications can bind to a socket and listen for incoming communications or use the socket to connect to other, remote services. But even on a single Linux system, sockets can be used to facilitate the communication flows. There are two socket types that can be used for process communication: UNIX domain sockets and netlink sockets. Netlink sockets are specific to the Linux operating system and are quite low-level, resembling the ioctl() system call usage. UNIX domain sockets, on the other hand, are higher-level and more directly accessible by administrators, which is why we explain them here in more detail. We can distinguish between two UNIX domain socket definitions, as with pipes: unnamed sockets and named sockets. And like pipes, the distinction is in the path used to identify a socket. Named sockets are created on the regular filesystem, while unnamed sockets are part of the sockfs pseudo filesystem. Similarly, sockets can be queried through utilities such as lsof or through the /proc//fd listings. There is another distinction regarding UNIX domain sockets though, namely, the communication format that the UNIX domain socket allows. UNIX domain sockets can be created as datagram sockets (data sent to the socket retains its chunk size and format) or streaming sockets (data sent to the socket can be read in different-sized chunks). This has some repercussions for the SELinux policy rules. For SELinux, communicating over UNIX domain sockets requires both domains to have the proper communication privileges toward the socket file type (open, read, and write), depending on the direction of the communication.

Controlling process communications

145

Additionally, the sending (client) domain requires additional privileges toward the receiving (server) domain: • The connectto privilege in the unix_stream_socket class in the case of stream sockets • The sendto privilege in the unix_dgram_socket class in the case of datagram sockets As you can see, the privileges depend on the communication type used across the socket.

Understanding netlink sockets Another socket type that can be used for process communication is netlink. Netlink sockets are sockets that allow user space applications to communicate and interact with kernel processes, and, in special cases (where network management is delegated to a user space process by the Linux kernel), also communicate with another user space application. Unlike the regular UNIX domain sockets, whose target context associates with the owner of that socket, netlink sockets are always local to the SELinux context. Put differently, when a domain such as sysadm_t wants to manipulate the kernel's routing information, it will open and communicate with the kernel through a netlink route socket, identified through the netlink_route_socket class: $ sesearch -s sysadm_t -t sysadm_t -c netlink_route_socket -A allow sysadm_t domain:netlink_route_socket getattr; allow sysadm_t sysadm_t:netlink_route_socket { append bind ... };

As applications gain more features, it might be that some of these features are no longer allowed by the current SELinux policy. Administrators will then need to update the SELinux policy to allow the netlink communication. An overview of supported netlink sockets can be devised from the netlink information on the manual page (man netlink), from which the SELinux classes can easily be derived. For instance, the NETLINK_XFRM socket is supported through the SELinux netlink_xfrm_socket class.

Dealing with TCP, UDP, and SCTP sockets When we go further up the chain, we look at socket communication over the network. In this case, rather than communicating directly between processes (and thus in Linux terminology between SELinux domains), the flows are from, and to, TCP, UDP, and Stream Control Transmission Protocol (SCTP) sockets.

146

Controlling Network Communications

SELinux will assign types to these ports as well, and these types are then the types to use for socket communication. For SELinux, a client application connecting to the DNS port (TCP port 53, which receives the dns_port_t type in most SELinux policies) uses the name_connect permission within the tcp_socket class toward the port type. The SCTP protocol (with the sctp_socket class) uses the same permission. For UDP services (and thus the udp_socket class), name_connect is not used. Daemon applications use the name_bind privileges to bind themselves to their associated port. Important note Support for SCTP has only been recently introduced in SELinux, and not all Linux distributions have updated their policies accordingly. To see whether SCTP support is active, check the value of the /sys/fs/selinux/ policy_capabilities/extended_socket_class file. A value of 1 means that the policy has SCTP support included, whereas a value of 0 (or an absent file) means that the system does not yet support SCTP.

Administrators can fine-tune which label to assign to which TCP, UDP, or SCTP port. For this, the semanage port command can be used. For instance, to list the current port definitions, you'd use this command: # semanage port -l SELinux Port Type Proto Port Number afs3_callback_port_t tcp 7001 ... http_port_t tcp 80, 81, 443, 488, 8008, 8009, ...

In this example, we see that the http_port_t label is assigned to a set of TCP ports. Web server domains that can bind to http_port_t are, as such, allowed to bind to any of the mentioned ports. To allow a daemon, such as an SSH server, to bind to other (or additional) ports, we need to tell SELinux to map this port to the appropriate label. For instance, to allow the SSH server to bind to port 10122, we first check whether this port already holds a dedicated label. This can be accomplished using the sepolicy command: $ sepolicy network -p 10122 10122: udp unreserved_port_t 1024-32767 10122: tcp unreserved_port_t 1024-32767 10122: sctp unreserved_port_t 1024-32767

Controlling process communications

147

The unreserved_port_t label is not a dedicated one, so we can assign the ssh_port_t label to it: # semanage port -a -t ssh_port_t -p tcp 10122

Removing a port definition works similarly: # semanage port -d -t ssh_port_t -p tcp 10122

When a specific port type is already assigned, then the utility will give the following error: # semanage port -a -t ssh_port_t -p tcp 80 ValueError: Port tcp/80 already defined

If this is the case and another port cannot be used, then no option exists other than to modify the SELinux policy.

Listing connection contexts Many of the tools in an administrator's arsenal can display security context information. As with the core utilities, most of these tools use the -Z option for this. For instance, to list the running network-bound services, netstat can be used: # netstat -naptZ | grep ':80' tcp 0 0 0.0.0.0:80 0.0.0.0:* LISTEN 17655/nginx: master system_u:system_r:httpd_t:s0

Even lsof displays the context when asked to: # lsof -i :80 -Z COMMAND PID SECURITY-CONTEXT USER FD TYPE DEVICE SIZE/OFF NODE NAME nginx 17655 system_u:system_r:httpd_t:s0 root 8u IPv4 31230 0t0 *:http (LISTEN)

Another advanced command for querying connections is the ss command. Just calling ss will display all the connections of the current system. When adding -Z, it adds the context information as well. For instance, the following command queries for listening TCP services: # ss -ltnZ

More advanced queries can be called as well — consult the ss manual page for more information.

148

Controlling Network Communications Note The use of the -Z option to show SELinux context information or consider SELinux context information in the activity that is requested by the user is a general but not mandatory practice amongst application developers. It is recommended to check the manual page of the application to confirm whether, and how, SELinux is supported by a tool. For instance, to get the ss manual page, run man ss.

All these interactions are still quite primitive in nature, with the last set (which focuses on sockets) being more network-related than the others. Once we look into interaction between systems, we might not have enough control through just the sockets though. To enable more fine-grained control, we'll look at firewall capabilities and their SECMARK support next.

Linux firewalling and SECMARK support The approach with TCP, UDP, and SCTP ports has a few downsides. One of them is that SELinux has no knowledge of the target host, so cannot reason about its security properties. This method also offers no way of limiting daemons from binding on any interface: in a multi-homed situation, we might want to make sure that a daemon only binds on the interface facing the internal network and not the internet-facing one, or vice versa. In the past, SELinux allowed support for this binding issue through the interface and node labels: a domain could be configured to only bind to one interface and not to any other, or even on a specific address (referred to as the node). This support had its flaws though, and has been largely deprecated in favor of SECMARK filtering. Before explaining SECMARK and how administrators can control it, let's first take a quick look at Linux's netfilter subsystem, the de facto standard for local firewall capabilities on Linux systems.

Introducing netfilter Like LSM, the Linux netfilter subsystem provides hooks in various stages of its networking stack processing framework, which can then be implemented by one or more modules. For instance, ip_tables (which uses the iptables command as its control application) is one of those modules, while ip6_tables and ebtables are other examples of netfilter modules. Modules implementing a netfilter hook must inform the netfilter framework of that hook's priority. This enables controllable ordering in the execution of modules (as multiple calls for the same hook can and will be used together).

Linux firewalling and SECMARK support

149

The ip_tables framework is the one we will be looking at in more detail because it supports the SECMARK approach. This framework is commonly referred to as just iptables, which is the name of its control application. We will be using this term for the remainder of this book. iptables offers several tables, functionally-oriented classifications for network processing. The common ones are as follows:

• The filter table enables the standard network-filtering capabilities. • The nat table is intended to modify routing-related information from packets, such as the source and/or destination address. • The mangle table is used to modify most of a packet's fields. • The raw table is enabled when administrators want to opt out certain packets/flows from the connection-tracking capabilities of netfilter. • The security table is offered to allow administrators to label packets once regular processing is complete. Within each table, iptables offers a default set of chains. These default chains specify where in the processing flow (and thus which hook in the netfilter framework) rules are to be processed. Each chain has a default policy – the default return value if none of the rules in a chain match. Within the chain, administrators can add several rules to process sequentially. When a rule matches, the configured action applies. This action can be to allow the packet to flow through this hook in the netfilter framework, be denied, or perform additional processing. Commonly provided chains (not all chains are offered for all tables) include the following: • The PREROUTING chain, which is the first packet-processing step once a packet is received • The INPUT chain, which is for processing packets meant for the local system • The FORWARD chain, which is for processing packets meant to be forwarded to another remote system • The OUTPUT chain, which is for processing packets originating from the local system • The POSTROUTING chain, which is the last packet-processing step before a packet is sent

150

Controlling Network Communications

Overly simplified, the implementation of these tables and their chains roughly associates with the priority of the calls within the netfilter framework. The chains are easily associated with the hooks provided by the netfilter framework, whereas the table tells netfilter which chain implementations are to be executed first.

Implementing security markings With packet labeling, we can use the filtering capabilities of iptables (and ip6tables) to assign labels to packets and connections. The idea is that the local firewall tags packets and connections and then the kernel uses SELinux to grant (or deny) application domains the right to use those tagged packets and connections. This packet labeling is known as SECurity MARKings (SECMARK). Although we use the term SECMARK, the framework consists of two markings: one for packets (SECMARK) and one for connections, that is, CONNection MARKings (CONNMARK). The SECMARK capabilities are offered through two tables, mangle and security. Only these tables currently have the action of tagging packets and connections available in their rule set: • The mangle table has a higher execution priority than most other tables. Implementing SECMARK rules on this level is generally done when all packets need to be labeled, even when many of these packets will eventually be dropped. • The security table is next in execution priority after the filter table. This allows the regular firewall rules to be executed first, and only tag those packets allowed by the regular firewall. Using the security table allows the filter table to implement the discretionary access control rules first and have SELinux execute its mandatory access control logic only if the DAC rules are executed successfully. Once a SECMARK action triggers, it will assign a packet type to the packet or communication. SELinux policy rules will then validate whether a domain is allowed to receive (recv) or send packets of a given type. For instance, the Firefox application (running in the mozilla_t domain) will be allowed to send and receive HTTP client packets: allow mozilla_t http_client_packet_t : packet { send recv };

Another supported permission set for SECMARK-related packets is forward_in and forward_out. These permissions are checked when using forwarding in netfilter.

Linux firewalling and SECMARK support

151

One important thing to be aware of is that once a SECMARK action is defined, then all the packets that eventually reach the operating system's applications will have a label associated with them — even if no SECMARK rule exists for the packet or connection that the kernel is inspecting. If that occurs, then the kernel applies the default unlabeled_t label. The default SELinux policy implemented in some distributions (such as CentOS) allows all domains to send and receive unlabeled_t packets, but this is not true for all Linux distributions.

Assigning labels to packets When no SECMARK-related rules are loaded in the netfilter subsystem, then SECMARK is not enabled and none of the SELinux rules related to SECMARK permissions are checked. The network packets are not labeled, so no enforcement can be applied to them. Of course, the regular socket-related access controls still apply — SECMARK is just an additional control measure. Once a single SECMARK rule is active, SELinux starts enforcing the packet-label mechanism on all packets. This means that all the network packets now need a label on them (as SELinux can only deal with labeled resources). The default label (the initial security context) for packets is unlabeled_t, which means that no marking rule matches this network packet. Because SECMARK rules are now enforced, SELinux checks all domains that interact with network packets to see whether they are authorized to send or receive these packets. To simplify management, some distributions enable send and receive rights against the unlabeled_t packets for all domains. Without these rules, all network services would stop functioning properly the moment a single SECMARK rule becomes active. To assign a label to a packet, we need to define a set of rules that match a particular network flow, and then call the SECMARK logic (to tag the packet or communication with a label). Most rules will immediately match the ACCEPT target as well, to allow this particular communication to reach the system. Let's implement two rules: • The first is to allow communication toward websites (port 80) and tag the related network packets with the http_client_packet_t type (so that web browsers are allowed to send and receive these packets). • The second is to allow communication toward the locally running web server (port 80 as well) and tag its related network packets with the http_server_packet_t type (so that web servers are allowed to send and receive these packets).

152

Controlling Network Communications

For each rule set, we also enable connection tracking so that related packets are automatically labeled correctly and passed. Use the following commands for the web server traffic: # iptables -t filter -A INPUT -m conntrack --ctstate ESTABLISHED,RELATED -j ACCEPT # iptables -t filter -A INPUT -p tcp -d 192.168.100.15 --dport 80 -j ACCEPT # iptables -t security -A INPUT -p tcp --dport 80 -j SECMARK --selctx "system_u:object_r:http_server_packet_t:s0" # iptables -t security -A INPUT -p tcp --dport 80 -j CONNSECMARK --save

Use these commands for the browser traffic: # iptables -t filter -A OUTPUT -m conntrack --ctstate ESTABLISHED -j ACCEPT # iptables -t filter -A OUTPUT -p tcp --dport 80 -j ACCEPT # iptables -t security -A OUTPUT -p tcp --dport 80 -j SECMARK --selctx "system_u:object_r:http_client_packet_t:s0" # iptables -t security -A OUTPUT -p tcp --dport 80 -j CONNSECMARK --save

Finally, to copy connection labels to the established and related packets, use the following commands: # iptables -t security ESTABLISHED,RELATED -j # iptables -t security ESTABLISHED,RELATED -j

-A INPUT -m state --state CONNSECMARK --restore -A OUTPUT -m state --state CONNSECMARK --restore

Even this simple example shows that firewall rule definitions are an art by themselves, and that the SECMARK labeling is just a small part of it. However, using the SECMARK rules makes it possible to allow certain traffic while still ensuring that only well-defined domains can interact with that traffic. For instance, it can be implemented on kiosk systems to only allow one browser to communicate with the internet while all other browsers and commands aren't. Tag all browsing-related traffic with a specific label, and only allow that browser domain the send and recv permissions on that label.

Linux firewalling and SECMARK support

153

Transitioning to nftables While iptables is still one of the most widely used firewall technologies on Linux, two other contenders (nftables and bpfilter) are rising rapidly in terms of popularity. The first of these, nftables, has a few operational benefits over iptables, while retaining focus on the netfilter support in the Linux kernel: • The code base for nftables and its Linux kernel support is much more streamlined. • Error reporting is much better. • Filtering rules can be incrementally changed rather than requiring a full reload of all rules. The nftables framework has recently received support for SECMARK, so let's see how to apply the http_server_packet_t and http_client_packet_t labels to the appropriate traffic. The most common approach for applying somewhat larger nftables rules is to use a configuration file with the nft interpreter set: #!/usr/sbin/nft -f flush ruleset table inet filter { secmark http_server { "system_u:object_r:http_server_packet_t:s0" } secmark http_client { "system_u:object_r:http_client_packet_t:s0" } map secmapping_in { type inet_service : secmark elements = { 80 : "http_server" } } map secmapping_out { type inet_service : secmark elements = { 80 : "http_client" } } chain input { type filter hook input priority 0; ct state new meta secmark set tcp dport map @secmapping_in ct state new ct secmark set meta secmark ct state established,related meta secmark set ct secmark } chain output {

154

Controlling Network Communications type filter hook output priority 0; ct state new meta secmark set tcp dport map @secmapping_out ct state new ct secmark set meta secmark ct state established,related meta secmark set ct secmark }

}

The syntax that nftables uses is recognizable when we compare it with iptables. The script starts with defining the SECMARK values. After that, we create a mapping between a port (80 in the example) and the value used for the SECMARK support. Of course, already established sessions also receive the appropriate SECMARK labeling. If we define multiple entries, the elements variable uses commas to separate the various values: elements = { 53 : "dns_client" , 80 : "http_client" , 443 : "http_client" }

Next to nftables. A second firewall solution that is gaining traction is eBPF, which we cover next.

Assessing eBPF eBPF (and the bpfilter command) is completely different in nature compared to iptables and nftables, so let's first see how eBPF functions before we cover the SELinux support details for it.

Understanding how eBPF works The extended Berkeley Packet Filter (eBPF) is a framework that uses an in-kernel virtual machine that interprets and executes eBPF code, rather low-level instructions comparable to processor instruction set operations. Because of its very low-level, yet processor-agnostic language, it can be used to create very fast, highly optimized rules. BPF was originally used for analyzing and filtering network traffic (for example, within tcpdump). Because of its high efficiency, it was soon found in other tools as well, growing beyond the plain network filtering and analysis capabilities. As BPF expanded toward other use cases, it became extended BPF, or eBPF. The eBPF framework in the Linux kernel has been successfully used for performance monitoring, where eBPF applications hook into runtime processes and kernel subsystems to measure performance and feed back the metrics to user-space applications. It, of course, also supports filtering on (network) sockets, cgroups, process scheduling, and many more — and the list is growing rapidly.

Linux firewalling and SECMARK support

155

As with the LSM framework, which uses hooks into the system calls and other security-sensitive operations in the Linux kernel, eBPF hooks into the Linux kernel as well. Occasionally it can use existing hooks (as with the Linux kernel probes or kprobes framework) and thus benefit from the stability of these interfaces. We can thus expect eBPF to grow its support further in other areas of the Linux kernel as well. eBPF applications (eBPF programs) are defined in user space, and then submitted to the Linux kernel. The kernel verifies the security and consistency of the code to ensure that the virtual machine will not attempt to break out of the boundaries it works in. If approved (possibly after the code is slightly altered, as the Linux kernel has some operations that modify eBPF code to suit the environment or security rules), the eBPF program runs in the Linux kernel (within its virtual machine) and executes its purpose. Note The Linux kernel can compile the eBPF instructions into native, processorspecific instructions, rather than having the virtual machine interpret them. However, as this leads to a higher security risk, this Just-In-Time (JIT) eBPF support is sometimes disabled by Linux distributions in their Linux kernels. It can be enabled by setting /proc/sys/net/core/bpf_jit_enable to 1.

These programs can load and save information in memory, called maps. These eBPF maps can be read or written to by user-space applications, and thus offer the main interface to interact with running eBPF programs. These maps are accessed through file descriptors, allowing processes to pass along and clone these file descriptors as needed. Various products and projects are using eBPF to create high-performance network capabilities, such as software-defined network configurations, DDoS mitigation rules, load balancers, and more. Unlike the netfilter-based firewalls, which rely on a massive code base within the kernel tuned through configuration, eBPF programs are built specifically for their purpose and nothing more, and only that code is actively running.

Securing eBPF programs and maps The default security measures in place for eBPF programs and maps are very limited, partly because lots of trust is put in the Linux kernel verifier (which verifies the eBPF code before it passes the code on to the virtual machine), and partly because the eBPF code was only allowed to be loaded when the process involved has the CAP_SYS_ADMIN capability. And as this capability basically means full system access, additional security controls were not deemed necessary.

156

Controlling Network Communications

Since Linux kernel 4.4, some types of eBPF programs (such as socket filtering) can be loaded even by unprivileged processes (but, of course, only toward the sockets these processes have access to). The system allows loading programs to work on cgroups socket buffers (skb) if the process has the CAP_NET_ADMIN capability. Recently, the permission to load eBPF programs has been added to the CAP_BPF and CAP_TRACING capabilities, although not all Linux distributions offer a Linux kernel that supports these capabilities already. But Linux administrators that want more fine-grained control over eBPF can use SELinux to tune and tweak eBPF handling. SELinux has a bpf class, which governs the basic eBPF operations: prog_load, prog_run, map_create, map_read, and map_write. Whenever a process creates a program or map, this program or map inherits the SELinux label of this process. If the file descriptors regarding these maps or programs are leaked, the malicious application still requires the necessary privileges toward this label before it can exploit it. User-space operations can interact with the eBPF framework through the /sys/fs/bpf virtual filesystem, so some Linux distributions associate a specific SELinux label (bpf_t) with this location as well. This allows administrators to manage access through SELinux policy rules in relation to this type. While eBPF is extremely extensible, the number of simplified frameworks surrounding it is small given its very early phase. We can, however, expect that more elaborate support will come soon, as a new tool called bpfilter is showing off the capabilities of eBPF-based firewalling on Linux systems.

Filtering traffic with bpfilter The bpfilter application is an application that builds a new eBPF program to filter and process traffic. It allows administrators to build firewall capabilities without understanding the low-level eBPF instructions, and has recently started supporting iptables: administrators create rules with iptables, and bpfilter translates and converts these into eBPF programs. Important note While bpfilter is part of the Linux kernel tree, it should be considered a proof-of-value currently, rather than a production-ready firewall capability.

bpfilter creates eBPF programs that hook inside the Linux kernel between the network device driver and the TCP/IP stack in a layer called the eXpress Data Path (XDP). At this level, the eBPF programs have access to the full network packet information (including link layer protocols such as Ethernet).

Securing high-speed InfiniBand networks

157

To use bpfilter, the Linux kernel needs to be built with the appropriate settings, including CONFIG_BPFILTER and CONFIG_BPFILTER_UMH. The latter is the bpfilter user mode helper that will capture iptables-generated firewall rules, and translate those into eBPF applications. Before we load the bpfilter user mode helper, we need to allow execmem permission in SELinux: # setsebool allow_execmem on

Next, load the bpfilter module, which will have the user mode helper active on the system: # modprobe bpfilter # dmesg | tail ... bpfilter: Loaded bpfilter_umh pid 2109

Now, load the iptables firewall using the commands listed previously. The instructions are translated into eBPF programs, as shown with bpftool: # bpftool p 1: xdp tag 8ec94a061de28c09 dev ens3 loaded_at Apr 25/23:19 uid:0 xlated 533B jited 943B memlock 4096B

The eBPF code itself can be displayed as well, but is hardly readable at this point for administrators. All of the aforementioned firewall capabilities interact with the TCP/IP stack supported within the Linux kernel. There are, however, networks that do not rely on TCP/IP, such as InfiniBand. Luckily, even on those more specialized network environments, SELinux can be used to control communication flows.

Securing high-speed InfiniBand networks The InfiniBand standard is a relatively recent (in network history) technology that enables very high throughput and very low latency. It accomplishes this by having a very low overhead on the network layer (protocol) and direct access from user applications to the network level. This direct access also has implications for SELinux, as the Linux kernel is no longer actively involved in the transport of data across an InfiniBand link. Let's first look at what InfiniBand looks like, after which we can see how to still apply SELinux controls to its communication flows.

158

Controlling Network Communications

Directly accessing memory One of the main premises of InfiniBand is to allow user applications to have direct access to the network. By itself, InfiniBand is a popular Remote Direct Memory Access (RDMA) implementation, which has received significant support from vendors. We find RDMA actively used in high-performance clusters. Because of the direct access, controls are only possible while setting up the access approach. Without SELinux, all that is needed to set up and manage InfiniBand communications is to have access to the device file itself. If a process can write to the InfiniBand device, then it can use InfiniBand. By default, these devices are only accessible by the root user. The InfiniBand devices are the network cards or Host Channel Adapters (HCA) and can have multiple ports. An InfiniBand port is the link or interface that connects to an InfiniBand subnet. The subnet is the high-speed network on which multiple machines (ports) are connected. As with regular networks, InfiniBand switches are used to facilitate communication across a subnet, and routers can be used to connect different subnets with each other. An InfiniBand subnet is managed by a Subnet Manager (SM). This is a process that coordinates the management of the different ports within the subnet, as well as the partitions. Partitions in InfiniBand are a way to differentiate between different communications within a subnet, like Virtual Local Area Networks (VLANs) in more regular networks. With partitioned communication, it is the subnet manager that tells which ports can be used for which partitions of the communication.

Protecting InfiniBand networks Unlike regular networks, where firewalls and switch-level access controls are the norm for preventing unauthorized access, InfiniBand has few protection measures in place. InfiniBand largely assumes that the network is within a trusted environment. However, that does not exclude us from applying more rigid controls over which process can access the InfiniBand network in SELinux. As the communication flow itself is directly mapped in-memory toward the devices, the Linux kernel does not have any hooks available to do packet-level controls like it can with regular TCP/UDP traffic (using the SECMARK capabilities), or even session-level controls with sockets. Instead, SELinux focuses on two main controls, as visualized in the following diagram:

Securing high-speed InfiniBand networks

159

Figure 5.1 – SELinux InfiniBand controls

These two main controls are as follows: • Controlling who can manage the InfiniBand subnet • Controlling who can access an InfiniBand partition To properly govern these controls, the semanage application assigns the right type to the appropriate InfiniBand resource. However, not all SELinux policies already contain the appropriate types, so we need to add those in as well.

Managing the InfiniBand subnet Let's start with managing the InfiniBand network. With InfiniBand on Linux, this is most often accomplished using the opensm application. Many InfiniBand adapters have multiple ports, allowing a server to participate in multiple InfiniBand subnets. With SELinux, we can control which domain can manage a subnet by controlling access to the InfiniBand port on a device. First, we need to assign a label to the InfiniBand port associated with a subnet. To accomplish that, we first need to obtain the right InfiniBand device, create the appropriate label (type), and then assign it to the port.

160

Controlling Network Communications

Let's start by querying the available InfiniBand-capable devices on the system using ibv_ devinfo: # ibv_devinfo hca_id: rxe0 transport: InfiniBand (0) fw_ver: 0.0.0 ... phys_port_cnt: port: 1 state: PORT_ACTIVE (4) ...

Next, we create a type (label) to assign to the port. This type is only used to validate the access from the opensm application to this port. We use the CIL language for this (which we will elaborate upon in Chapter 16, Developing Policies with SELinux CIL). Create a file with the following content (let's call it infiniband_subnet.cil): (typeattribute ibendport_type) (type local_ibendport_t) (typeattributeset ibendport_type local_ibendport_t) (allow opensm_t local_ibendport_t (infiniband_endport (manage_ subnet)))

In the previous code, we enhance the SELinux policy with a new type called local_ ibendport_t, assign it the ibendport_type attribute, and then grant the opensm_t domain the manage_subnet privilege within the infiniband_endport class. Let's load this policy enhancement: # semodule -i infiniband_subnet.cil

Finally, we assign this newly created type to the InfiniBand port: # semanage ibendport -a -t local_ibendport_t -z rxe0 1

This command assigns the local_ibendport_t type to port number 1 of the rxe0 device (as obtained from ibv_devinfo). Once this mapping is in place, we can query it using semanage as well: # semanage ibendport -l SELinux IB End Port Type IB Device Name Port Number local_ibendport_t rxe0 0x1

Without any mappings, the command does not display any output.

Securing high-speed InfiniBand networks

161

Important note Currently, most Linux distributions have not incorporated InfiniBand support within the SELinux policy, requiring us to create our own custom labels. We can expect that distributions will add in default types for InfiniBand resources, and that SELinux support for InfiniBand will be extended with sane defaults.

If we use InfiniBand on an SELinux-enabled system without any port mappings, the initial security context for unlabeled classes will be used as the label for this port, namely, unlabeled_t. It is, however, not recommended to stick to this label, as it is more widely used for unlabeled resources. Granting any privilege to the unlabeled_t type should be limited to highly privileged processes, and its use should be carefully considered to ensure that logging interpretation and SELinux policy rules vis-à-vis InfiniBand resources are clear (through well-documented types).

Controlling access to InfiniBand partitions While the previous section focused on allowing the management application opensm to manage a subnet, this section will focus on restricting access to the InfiniBand network to the right domains. As mentioned before, an InfiniBand subnet can be divided further into separate networks using InfiniBand partitions. Originally, these partitions are used to allow Quality of Service (QoS) or specific bandwidth and performance requirements on flows. The SM defines the partitions and its attributes, and applications use a Partition Key (P_Key) to inform the InfiniBand network as regards to which partition certain communications must be done. SELinux can govern these partitions by creating a mapping between the InfiniBand subnet plus P_Key and an SELinux type. However, as with the subnet management, we need to find the appropriate details first and create an appropriate SELinux type before we can define the mapping. Let's start by figuring out the subnet and partition details. Both are managed by opensm. If you do not have access to the opensm configuration, then you need to ascertain these details from the (InfiniBand) network administrator. Within the opensm partition configuration (/etc/rdma/partitions.conf), the subnet and prefix can be found as follows: # grep '=0x' /etc/rdma/partitions.conf Default=0x7fff, rate=3, mtu=4, scope=2, defmember=full; Default=0x7fff, ipoib, rate=3, mtu=4, scope=2; rxe0_1=0x0610, rate=7, mtu=4, scope=2, defmember=full; rxe0_1=0x0610, ipoib, rate=7, mtu=4, scope=2;

162

Controlling Network Communications

In this example, two partitions are defined. The first one is the default partition, which needs to remain (0x7fff). The second partition with key 0x0610 is active on the rxe0 device and port 1. It is this second partition that we will protect with SELinux. Let's create a new type to assign to this partition. We use the CIL format again to define the policy enhancement, and store these rules in a file called infiniband_pkey.cil: (typeattribute ibpkey_type) (type local_ibpkey_t) (typeattributeset ibpkey_type local_ibpkey_t) (allow unconfined_t local_ibpkey_t (infiniband_pkey (access)))

Within this example, we've created the local_ibpkey_t type, assigned it to the ibpkey_type attribute, and granted unconfined_t access privilege within the infiniband_pkey class. Let's load the policy: # semodule -i infiniband_pkey.cil

We can now create an appropriate mapping to this partition, and limit it to the ff12:: subnet prefix: # semanage ibpkey -a -t local_ibpkey_t -x ff12:: 0x0610 # semanage ibpkey -l SELinux IB PKey Type Subnet_Prefix Pkey Number local_ibpkey_t ff12:: 0x610

While we can create separate types for each partition, we can also use an SELinux range to use SELinux category support: # semanage ibpkey -a -t local_ibpkey_t -r s0-s0:c0.c4 -x ff12:: 0x0610

With categories, we can grant access based on the source domain category, something we benefit from with other network protection measures such as labeled networking, which we tackle next.

Understanding labeled networking

163

Understanding labeled networking Another approach to further fine-tune access controls on the network level is to introduce labeled networking. With labeled networking, security information passes on between hosts (unlike SECMARK, which only starts when the netfilter subsystem receives the packet, and whose marking never leaves the host). This is also known as peer labeling, as the security information passes on between hosts (peers). The advantage of labeled networking is that security information remains across the network, allowing end-to-end enforcement on mandatory access-control settings between systems as well as retaining the sensitivity level of communication flows between systems. The major downside, however, is that this requires an additional network technology (protocol) that can manage labels on network packets or flows. SELinux currently supports two implementations as part of the labeled networking approach: NetLabel and labeled IPsec. With NetLabel, two implementations exist: fallback labeling and CIPSO. In both cases, only the sensitivity of the source domain is retained across the communication. Labeled IPsec supports transporting the entire security context with it. Note NetLabel actually supports loopback-enabled, full-label support. In that case, the full label (and not only the sensitivity and categories) is passed on. However, this only works for communications that go through the loopback interface and, as such, do not leave the current host.

Quite some time ago, support for NetLabel/CIPSO and labeled IPsec merged into a common framework, which introduces three additional privilege checks in SELinux: interface checking, node checking, and peer checking. These privilege checks are only active when labeled traffic is used; without labeled traffic, these checks are simply ignored.

Fallback labeling with NetLabel The NetLabel project supports fallback labeling, where administrators can assign labels to traffic from or to network locations that don't use labeled networking. By using fallback labeling, the peer controls mentioned in the next few sections can be applied even without labeled IPsec or NetLabel/CIPSO being in place. The netlabelctl command controls the NetLabel configurations. Let's create a fallback label assignment for all traffic originating from the 192.168.100.1 address: # netlabelctl unlbl add interface:eth0 address:192.168.100.1 label:system_u:object_r:netlabel_peer_t:s0

164

Controlling Network Communications

To list the current definitions, use the following command: # netlabelctl -p unlbl list Accept unlabeled packets : on Configured NetLabel address mappings (1) interface: eth0 address: 192.168.100.1/32 label: "system_u:object_r:netlabel_peer_t:s0"

With this rule in place, labeled networking is active. Any traffic originating from the 192.168.100.1 address will be labeled with the netlabel_peer_t:s0 label, while all other traffic will be labeled with the (default) unlabeled_t:s0 label. Of course, the SELinux policy must allow all domains to have the recv permission from either the unlabeled_t peers or the netlabel_peer_t peers. Fallback labeling is useful for supporting a mix of labeled networking environments and non-labeled networks, which is why we list it here before documenting the various labeled networking technologies.

Limiting flows based on the network interface The idea involving interface checking is that each packet that comes into a system passes an ingress check on an interface, whereas a packet that goes out of a system passes an egress check. ingress and egress are the SELinux permissions involved, whereas interfaces are given a security context. Interface labels can be granted using the semanage tool and are especially useful for assigning sensitivity levels to interfaces in case of MLS, although assigning different labels to the interface is also possible (but requires more adjustments to the running SELinux policy to return with a working system): # semanage interface -a -t netif_t -r s1-s1:c0.c128 eth0

Like the other semanage commands, we can view the current mappings as follows: # semanage interface -l SELinux Interface Context eth0 system_u:object_r:netif_t:s1-s1:c0.c128

Keep in mind that for inbound communications, the acting domain is the peer. With labeled IPsec, this would be the client domain initiating the connection, whereas in NetLabel/CIPSO, this is the associated peer label (such as netlabel_peer_t).

Understanding labeled networking

165

By default, the interface is labeled with netif_t and without sensitivity constraints. This will, however, not be shown in the semanage interface -l output as its default output is empty.

Accepting peer communication from selected hosts SELinux nodes represent specific hosts (or a network of hosts) that data is sent to (sendto) or received from (recvfrom) and are handled through the SELinux node class. Just like interfaces, these can be listed and defined by the semanage tool. In the following example, we mark the 10.0.0.0/8 network with the node_t type and associate a set of categories with it: # semanage node -a -t node_t -p ipv4 -M 255.255.255.255 -r s0-s0:c0.c128 192.168.100.1

Again, we can list the current definitions, too: # semanage node -l

Like the network interface flow, the acting domain for incoming communications is the peer label. By default, nodes are labeled with node_t and without category constraints. This will, however, not be shown in the semanage node -l output as its default output is empty.

Verifying peer-to-peer flow The final check is a peer class check. For labeled IPsec, this is the label of the socket sending out the data (such as mozilla_t). For NetLabel/CIPSO, however, the peer will be static, based on the source, as CIPSO is only able to pass on sensitivity levels. A common label seen for NetLabel is netlabel_peer_t. Unlike the interface and node checks, peer checks have the peer domain as the target rather than the source. Important note In all the labeled networking use cases, the process listed in a denial has nothing to do with the denial shown in the audit logs. This is because the denial triggers from within a kernel subsystem rather than through a call made by a user process. As a result, the kernel interrupts an unrelated process to prepare and log the denial, and this process name is used in the denial event.

166

Controlling Network Communications

To finish up, look at the following diagram, which provides an overview of these various controls and the level to which they apply:

Figure 5.2 – Schematic overview of the various network-related SELinux controls

The top-level controls are handled on the domain level (such as httpd_t), whereas the bottom-level controls are on the peer level (such as netlabel_peer_t).

Using old-style controls Most Linux distributions enable the network_peer_control capability. This is an enhancement within the SELinux subsystem that uses the previously mentioned peer class for verifying peer-to-peer flow. However, SELinux policies can opt to return to the previous approach, where peer-to-peer flow is no longer controlled over the peer class, but uses the tcp_socket class for communication. In that case, the tcp_socket class will be used against the peer domain, and it will also use the recvfrom permission (on top of the existing tcp_ socket permissions). The current value of the network_peer_control capability can be queried through the SELinux filesystem: # cat /sys/fs/selinux/policy_capabilities/network_peer_controls 1

Using labeled IPsec with SELinux

167

If the value is 0, then the previously mentioned peer controls will be handled through the tcp_socket class instead of the peer class. The default labeled networking controls within SELinux do not pass on any process context, and the use of fallback labeling with NetLabel is most commonly used in environments where the system participates in both labeled as well as unlabeled networks. However, there is a much more common networking implementation that not only supports labeled networking, but even passes on the domain context and does not require specialized environments: labeled IPsec.

Using labeled IPsec with SELinux Although setting up and maintaining an IPsec setup is far beyond the scope of this book, let's look at a simple IPsec example to show how to enable labeled IPsec on a system. Remember that the labeled network controls on the interface, node, and peer levels, as mentioned earlier, are automatically enabled the moment we use labeled IPsec. In an IPsec setup, there are three important concepts to be aware of: • The security policy database (SPD) contains the rules and information for the kernel to know when communication should be handled by an IP policy (and, as a result, handled through a security association). • A security association (SA) is a one-way channel between two hosts and contains all the security information about the channel. When labeled IPsec is in use, it also contains the context information of the client that caused the security association to materialize. • The security association database (SAD) contains the individual security associations. Security associations with a labeled IPsec setup are no longer purely indexed by the source and target address, but also the source context. As such, a Linux system that participates in a labeled IPsec setup will easily have several dozen SAs for a single communication flow between hosts, as each SA now also represents a client domain.

168

Controlling Network Communications

Labeled IPsec introduces a few additional access controls through SELinux: • Individual entries in the SPD are given a context. Domains that want to obtain an SA need to have the polmatch privilege (part of the association class) against this context. Also, domains that initiate an SA need to have the setcontext privilege (also part of the association class) against the target domain. • Only authorized domains can make modifications to the SPD, which is also governed through the setcontext privilege, but now also against the SPD context entries. This privilege is generally granted to IPsec tools, such as Libreswan's pluto (ipsec_t). • Domains that participate in IPsec communication must have the sendto privilege with their own association and the recvfrom privilege with the association of the peer domain. The receiving domain also requires the recv privilege from the peer class associated with the peer domain. So while labeled IPsec cannot govern whether mozilla_t can communicate with httpd_t (as mozilla_t only needs to be able to send to its own association), it can control whether httpd_t allows or denies incoming communication from mozilla_t (as it requires the recvfrom privilege on the mozilla_t association). The following diagram displays this complex game of privileges:

Figure 5.3 – Example SELinux controls for labeled IPsec

In the next example, we will set up a simple IPsec tunnel between two hosts using the Libreswan tool.

Using labeled IPsec with SELinux

169

Setting up regular IPsec Configuring Libreswan is a matter of configuring Libreswan's main configuration file (ipsec.conf). Most distributions will use an include directory (such as /etc/ipsec.d) where admins or applications can place connection-specific settings. Generally, this include directory is used for the actual IPsec configurations, whereas the general ipsec.conf file is for Libreswan behavior. To create a host-to-host connection, we first define a shared secret on both hosts. Let's call the connection rem1-rem2 (as those are the hostnames used for the two hosts), so the shared secret will be stored as /etc/ipsec.d/rem1-rem2.secrets: 192.168.100.4 192.168.100.5 : PSK "somesharedkey"

Next, we define the VPN connection in /etc/ipsec.d/rem1-rem2.conf as follows: conn rem1-rem2 left=192.168.100.4 right=192.168.100.5 auto=start authby=secret #labeled-ipsec=yes #policy-label=system_u:object_r:ipsec_spd_t:s0

The settings that enable labeled IPsec are commented out for now to first test the IPsec connection without this feature. Launch the IPsec service on both systems: # systemctl start ipsec

Verify whether the connection works, for instance, by checking the network traffic with tcpdump, or by checking the state with ip xfrm state.

Enabling labeled IPsec To use labeled IPsec with Libreswan, uncomment the labeled-ipsec and policylabel directives in the /etc/ipsec.d/rem1-rem2.conf IPsec definition. Restart the ipsec service, and try the connection again.

170

Controlling Network Communications

When an application tries to communicate over IPsec with remote domains, pluto (or any other Internet Key Exchange version 2 (IKEv2) client that supports labeled IPsec) will exchange the necessary information (including context) with the other side. Both sides will then update the SPD with the necessary SAs and associate the same security policy information (SPI) with it. From that point onward, the sending side will add the agreed-upon SPI information to the IPsec packets so that the remote side can immediately associate the right context with it again. The huge advantage here is that the client and server contexts, including sensitivity and categories, are synchronized (they are not actually sent over the wire with each packet, but exchanged initially when the security associations are set up). In certain specialized or highly secure environments, labeled networking is supported within the network itself. The most common labeling technology used is CIPSO, whose SELinux support we cover next.

Supporting CIPSO with NetLabel and SELinux NetLabel/CIPSO labels and transmits sensitivities across the network. Unlike labeled IPsec, no other context information is sent or synchronized. So, when we consider the communication flows between two points, they will have a default, common SELinux type (rather than the SELinux type associated with the source or target) but will have sensitivity labels based on the sensitivity label of the remote side. Part of NetLabel's configuration are mapping definitions that inform the system which communication flows (from selected interfaces, or even from configured IP addresses) are for a certain Domain of Interpretation (DOI). The CIPSO standard defines the DOI as a collection of systems that interpret the CIPSO label similarly, or, in our case, use the same SELinux policy and configuration of sensitivity labels. Once these mappings have been established, NetLabel/CIPSO will pass on the sensitivity information (and categories) between hosts. The context we will see on the communication flows will be netlabel_peer_t, a default context assigned to NetLabel/CIPSO-originated traffic. Through this approach, we can start daemons with a sensitivity range and thus only accept connections from users or clients that have the right security clearance, even on remote, NetLabel/CIPSO-enabled systems.

Supporting CIPSO with NetLabel and SELinux

171

Configuring CIPSO mappings A preliminary requirement for having a good CIPSO-enabled network is to have a common understanding of which DOI will be used and what its consequences are. Labeled networks can use different DOIs for specific purposes. Along with the DOI, we also need to take care of how the categories and sensitivities are passed on over the CIPSO-enabled network. The CIPSO tag controls this setting, and NetLabel supports this with the following three values: • With tags:1, the categories are provided in the CIPSO package in a bitmap approach. This is the most common approach, but limits the number of supported categories to 240 (from 0 to 239). • With tags:2, the categories are enumerated separately. This allows a wider range of categories (up to 65,543), but only supports at most 15 enumerated categories. Try to use tags:2 when you have many categories but for each scope, only a few categories need to be supported. • With tags:5, the categories can be mentioned in a ranged approach (lowest and highest), with at most seven such low/high pairs. Note that the CIPSO tag results are handled under the hood: system administrators only need to configure the NetLabel mapping to use a selected tag value. Let's assume that we have two CIPSO-enabled networks, which have 10.1.0.0/16 associated with doi:1 and 10.2.0.0/16 associated with doi:2. Both use the tag value 1. First, we enable CIPSO and allow it to pass CIPSO-labeled packages with the DOI set to either 1 or 2. We don't perform any translations (so the category and sensitivity set on the CIPSO package is the one used by SELinux): # netlabelctl cipsov4 add pass doi:1 tags:1 # netlabelctl cipsov4 add pass doi:2 tags:1

If we need to translate (say that we use sensitivity s0-s3 while the CIPSO network uses sensitivity 100-103), a command would look like so: # netlabelctl cipsov4 add std doi:1 tags:1 levels:0=100,1=101,2=102

172

Controlling Network Communications

Next, we implement mapping rules, telling the NetLabel configuration which network traffic is to be associated with doi:1 or doi:2: # netlabelctl map del default # netlabelctl map add default address:10.1.0.0/16 protocol:cipsov4,1 # netlabelctl map add default address:10.2.0.0/16 protocol:cipsov4,2

To list the current mappings, use the list option: # netlabelctl map list -p Configured NetLabel domain mappings (2) domain: DEFAULT (IPv4) address: 10.1.0.0/16 protocol: CIPSO, DOI = 1 domain: DEFAULT (IPv4) address: 10.2.0.0/16 protocol: CIPSO, DOI = 2

That's it. We removed the initial default mapping (as that would prevent the addition of new default mappings) and then configured NetLabel to tag traffic for the given networks with the right CIPSO configuration.

Adding domain-specific mappings NetLabel can also be configured to ensure that given SELinux domains use a well-defined DOI rather than the default one configured earlier on. For instance, to have the SSH daemon (running in the sshd_t domain) have its network traffic labeled with CIPSO doi:3, we'd use this: # netlabelctl cipsov4 add pass doi:3 tags:1 # netlabelctl map add domain:sshd_t protocol:cipsov4,3

The mapping rules can even be more selective than that. We can tell NetLabel to use doi:2 for SSH traffic originating from one network, use doi:3 for SSH traffic originating from another network, and even use unlabeled network traffic when it comes from any other network: # netlabelctl map del domain:sshd_t protocol:cipsov4,3 # netlabelctl map add domain:sshd_t address:10.1.0.0/16 protocol:cipsov4,1

Supporting CIPSO with NetLabel and SELinux

173

# netlabelctl map add domain:sshd_t address:10.4.0.0/16 protocol:cipsov4,3 # netlabelctl map add domain:sshd_t address:0.0.0.0/0 protocol:unlbl

The NetLabel framework will try to match the most specific rule first, so 0.0.0.0/0 is only matched when no other rule matches.

Using local CIPSO definitions As mentioned before, NetLabel, by default, only passes the sensitivity and categories. However, when using local (over the loopback interface) CIPSO, it is possible to use full label controls. When enabled, peer controls will not be applied against the default netlabel_peer_t type, but will use the client or server domain. To use local CIPSO definitions, first declare the DOI for local use: # netlabelctl cipsov4 add local doi:5

Next, have the local communication use the defined DOI (5 in our example): # netlabelctl map add default address:127.0.0.1 protocol:cipsov4,5

With this enabled, local communication will be associated with doi:5 and use the local mapping, passing the full label to the mandatory access control system (SELinux).

Supporting IPv6 CALIPSO CIPSO is an IPv4 protocol, but a similar framework exists for IPv6, named Common Architecture Label IPv6 Security Option (CALIPSO). As with CIPSO, CALIPSO is supported by the NetLabel project. When we need CALIPSO support, the protocol target is calipso rather than cipsov4. CALIPSO has a few small differences compared to CIPSO in NetLabel: • Only one tag type is supported (unlike CIPSO's three tag types). As a result, CALIPSO administrators do not need to specify tags:# anywhere. • CALIPSO only uses pass-through mode. Translations are not supported. • The NetLabel CALIPSO implementation currently does not support local mode, where the full label would be passed on. Beyond these differences, the use of CALIPSO is very similar to CIPSO.

174

Controlling Network Communications

Summary SELinux, by default, uses access controls based on the file representation of communication primitives or the sockets used. On InfiniBand networks, access controls are limited to accessing the InfiniBand port and partitions. For TCP, UDP, and SCTP ports, administrators have some leeway in handling the controls through the semanage command without resorting to SELinux policy updates. Once we go into the realms of network-based communication, more advanced communication control can be accomplished through Linux netfilter support, using SECMARK labeling, and through peer labeling. In the case of SECMARK labeling, local firewall rules are used to map contexts to packets, which are then governed through SELinux policy. With peer labeling, either the application context itself (labeled IPsec) or its sensitivity level (netfilter/CIPSO) identify the resources the access controls apply. This allows an almost application-to-application network flow control through SELinux policies. We learned that the most common firewall frameworks (iptables and nftables) support SECMARK already, while the more recent eBPF-based bpfilter application has yet to receive this support. In the next chapter, we look at how we can use common infrastructure-as-code frameworks to address the various SELinux controls in a server environment.

Questions 1. How do you map an SELinux type to a TCP port? 2. Does SECMARK labeling change the network packets as they go over the wire? 3. What semanage subcommands are used for InfiniBand support? 4. Is specialized equipment needed for labeled IPsec?

6

Configuring SELinux through Infrastructure-asCode Orchestration With the advent of large distributed application platforms, cloud services, and the high adoption of virtualized infrastructure, system administrators are actively managing their systems through Infrastructure-as-Code frameworks: orchestration and configuration tooling that uses source code-like information to manage the systems. In this chapter, administrators will learn how to distribute and load custom SELinux policy modules, set file context definitions and apply those to the systems, set the permissive state of the system or SELinux domains, configure the SELinux settings on the systems, and how to customize SELinux actions if they are not supported by the tooling. We will apply this with four popular automation frameworks: Ansible, Chef, Puppet, and SaltStack.

176

Configuring SELinux through Infrastructure-as-Code Orchestration

We will cover the following topics in this chapter: • Introducing the target settings and policies • Using Ansible for SELinux system administration • Utilizing SaltStack to configure SELinux • Automating system management with Puppet • Wielding Chef for system automation

Technical requirements The code files for this chapter can be found in our Git repository at https://github. com/PacktPublishing/SELinux-System-Administration-ThirdEdition. Check out the following video to see the Code in Action: https://bit.ly/2T4Fksv

Introducing the target settings and policies Before we embark on the journey of using these four automation frameworks, we need to clarify what we want to accomplish. After all, to truly compare automation frameworks, we need to test each framework with the same tests each time.

The idempotency of actions Whenever we create a remote management environment with a central repository, we need to consider the impact of running remote management activities on the system. A very common best practice, strongly adopted by all these frameworks, is idempotency. An idempotent task is a task that will not modify a system if the system's state is already how it should be. Or, differently put, repeatedly executing a task does not affect the system or the processes that run on it if nothing needs to be changed. As an example, consider loading an SELinux module: if the module is already loaded, then the module should not be reloaded. If it isn't loaded yet, then we will load the proper module. While most actions supported by the automation frameworks are idempotent, we will need to create custom actions ourselves if the framework does not support what we want. For instance, if the framework does not support loading SELinux modules, then we need to write our own code to do so.

Introducing the target settings and policies

177

Most orchestration frameworks will envelope non-idempotent tasks in a definition that is more idempotent. For instance, if a change in a configuration file requires a system reboot, then the enveloped definition would be something like reboot after file change. The engine can check the state of the file (when it changed) and the system (when it rebooted) and deduce whether a reboot is needed or not, even though a system reboot as a task is a non-idempotent task.

Policy and state management The first set of scenarios that we want to support through the automation frameworks is to ensure that SELinux is active (enforcing) and that the right SELinux policy is loaded, a task usually performed by the machine's package management system. While allowing the package management system to handle this is convenient, it only offers the ability to use distribution-specific default policies. Sysadmins of systems with different security requirements will be restrained when using default policies and will need to create custom policies and policy handling routines. So, we will examine how to distribute and load custom policies. The custom policy we will use in the examples is a CIL policy, which is very new and often not directly supported by the automation frameworks. However, it gives us a nice reoccurring situation to create custom rules within the automation framework. We store the policy itself in a file called test.cil that has the following content: (auditallow staff_sudo_t sysadm_t (process (transition)))

This simple policy will enable logging any transition from the staff_sudo_t domain to the sysadm_t domain and is easy to test out with sudo. In our example, it serves no other purpose than to quickly allow us to verify that the policy has been correctly loaded. State-wise, we will ensure that the system is in enforcing mode, but have the zoneminder_t SELinux domain marked as permissive.

SELinux configuration settings The second group of actions we want to take up is to configure the system with the various SELinux settings we've discussed in different chapters before. Most of these we've seen through the semanage commands, and we will learn how the various automation frameworks support these entries, and to what extent.

178

Configuring SELinux through Infrastructure-as-Code Orchestration

We will not go through every setting, but rather focus on the supported configuration sets within each automation framework. If a framework does not support a particular configuration (such as the semanage ibpkey one, which is fairly new), we will need to create custom actions for this. In that case, we will show how to approach this once, as it is a recurring and similar approach for the others as well.

Setting file contexts The third and final group of actions we want to see is how the automation frameworks support applying file contexts to resources. This can be applying an semanage fcontext configuration, after which a restore operation is done (such as with restorecon), but also validating whether the framework supports applying contexts directly. Directly applying the context allows administrators to use the frameworks directly without having to twiddle with creating and reapplying file context definitions (which can have some performance overhead). However, this should only be considered if the automation framework is the sole method through which system changes can be made. In any other case, having missing file context definitions might lead to administrators resetting contexts to an incorrect state.

Recovering from mistakes In this chapter, we're diving into the various frameworks that allow managing SELinux across a multitude of systems. It is not the intention of this chapter to explain the frameworks themselves in detail, nor the secure configuration of the frameworks themselves. We don't recommend immediately applying this to production systems without testing first, and make sure to have backups! That being said, many settings applied in this chapter are easily corrected if things fail. We refer to Chapter 2, Understanding SELinux Decisions and Logging, to selectively put SELinux in permissive mode if needed. Furthermore, each framework can easily be suspended, allowing administrators to correct issues without being affected by the framework overwriting the changes immediately after.

Comparing frameworks Every framework we discuss further has its own approach to infrastructure automation and configuration. It is not the intention of this book to dwell on the details of each framework, but rather to focus on its core support and how it deals with SELinux. We will also abstract away how to handle different Linux distributions and have all examples based on CentOS.

Using Ansible for SELinux system administration

179

Furthermore, these frameworks are continuously improving and evolving. When we consider these frameworks in this chapter, we only explore how they are commonly used, and not how they can specialize in specific deployments. For instance, if a framework uses an agent-based architecture by default but also supports SSH-based connections, we will only consider the agent-based one in this book, as that is the default setup for these frameworks and we want to focus on the SELinux configuration support features. But don't let this stop you from experimenting with the frameworks further and adapting them to your own liking! That said, let's dive into our first engine, Ansible.

Using Ansible for SELinux system administration The first orchestration and automation tooling we'll consider is Ansible, a very popular open source solution for the remote management of systems. Ansible has commercial backing through Red Hat but does not limit its support to Red Hat or even Linux systems. Other environments such as Windows environments or even network setups have significant Ansible-based support.

How Ansible works Ansible generally uses a central server that hosts the configuration and interprets the settings. The Ansible runtime then connects to the remote systems over SSH, sending the necessary data to a temporary location, and then executes the steps locally. The use of SSH as its main connection approach has significant advantages: administrators know how this protocol works and how to configure and control it. Furthermore, Ansible does not require any additional deployments on the target machines, except for Python and libselinux's Python bindings (which are often installed on SELinux-enabled machines by default). Ansible knows how to address the various resources through its modules. Ansible modules contain the logic that Ansible uses to execute tasks correctly. The module code is distributed to the target machines and is executed on the remote systems. The definitions that administrators configure systems with are stored in Ansible playbooks. Playbooks define how a system should be configured, and Ansible will read and interpret playbooks to see what it must execute on each system.

180

Configuring SELinux through Infrastructure-as-Code Orchestration

To facilitate the management of Ansible playbooks in larger environments, Ansible uses Ansible roles to bundle coherent definitions. Administrators can then, in their playbooks, assign roles to systems to automatically uplift the state of those systems accordingly. For instance, a role can be created to create a properly configured web server, a database, and so on. In this chapter, we will create a role called packt_selinux and apply it to a remote system. Within that role, we will show how to configure and execute the various SELinux tasks using Ansible.

Installing and configuring Ansible To install and set up Ansible, most Linux distributions offer out-of-the-box support for the framework. On CentOS, the following steps can be taken. Users of other distributions can easily deduce the steps for their platform: 1. You need to enable Extra Packages for Enterprise Linux (EPEL), after which you can install Ansible easily. Execute this on the master node (from which you want to manage the other systems): # yum install epel-release # yum install ansible

2. Once installed, create an SSH key pair to use between the master system and the target systems that we will be managing with Ansible. Use the ssh-keygen command to create a key pair on the master system, and then copy the public key (~/.ssh/id_rsa.pub) to the remote systems, saving it as ~/.ssh/ authorized_keys: # ssh-keygen # scp ~/.ssh/id_rsa.pub rem1:/root/.ssh/authorized_keys

3. Test to see whether the remote connection works properly, for instance, by executing the id command remotely: # ssh rem1 id

4. If the test is successful, we can configure Ansible to see this remote system as one of the nodes it will be managing. To accomplish this, edit /etc/ansible/hosts and add the hostname to the list: # cat /etc/ansible/hosts rem1

Using Ansible for SELinux system administration

181

5. To see whether Ansible can correctly manage the remote system, we can ask it to gather all the facts about the remote system. Facts in Ansible represent the discovered settings of the remote system and can be used to fine-tune playbooks and roles later. For instance, the Ansible facts discovered of the distribution can be used to select which package name an installation uses: # ansible all -m setup

This instruction asks all managed hosts (all, reflecting all entries in /etc/ ansible/hosts) to execute the tasks in the setup module. The output of the last task is a large set of discovered facts, showing us that the connection succeeded and that Ansible is ready to manage the remote system.

Creating and testing the Ansible role To allow reusable configurations across multiple systems, Ansible recommends the use of its Ansible roles. We will create a role called packt_selinux, have it create a custom directory, and then assign this role to the remote system: 1. Use ansible-galaxy to create an empty yet ready-to-use role: # cd /etc/ansible/roles # ansible-galaxy init packt_selinux --offline - Role packt_selinux was created successfully

This command will create the necessary files and directories that constitute a role. The file we will use is packt_selinux/tasks/main.yml, which will host all the settings and definitions we want to apply when we assign the packt_selinux role to a system. The other directories are, for our brief introduction to Ansible, less relevant, but play an important role in making sufficiently modular roles. 2. Edit the main.yml file and have it create a custom directory. The content of the file should look like this: --- name: Create /usr/share/selinux/custom directory file: path: /usr/share/selinux/custom owner: root group: root mode: '0755' state: directory

182

Configuring SELinux through Infrastructure-as-Code Orchestration

In later steps, this file will be extended with more and more blocks. Each block will start with a name that identifies the block, and then the state definition. In the current block, we used Ansible's file module to assert that a file or directory is available with the parameters given. 3. Assign the role to the remote system and apply the playbook. We accomplish this by first creating an /etc/ansible/site.yml file with the following content: --- hosts: all roles: - packt_selinux

4. Run this playbook to apply the setting defined in our role to the remote systems: # ansible-playbook /etc/ansible/site.yml

Ansible will display its progress, as well as for which tasks it has executed a change. In our case, a change would mean that the directory has been created. Now that we have tested our role and assigned the role to the remote system, all we need to do is update the role gradually until it contains all the logic we need. No other configuration is needed, and after each change, we can rerun the ansible-playbook command from the main server.

Assigning SELinux contexts to filesystem resources with Ansible In the current role, we create a custom directory inside /usr/share/selinux. This parent directory has the usr_t SELinux type set, so the newly created subdirectory has it as well. The SELinux user of this directory, however, will be different, as Ansible has created the directory after remotely logging in to the system. In a default CentOS configuration, this means that the target directory's context will have unconfined_u as its SELinux user component. Let's update the definition in main.yml and explicitly set the SELinux user and type: --- name: Create /usr/share/selinux/custom directory file: path: /usr/share/selinux/custom owner: root group: root mode: '0755'

Using Ansible for SELinux system administration

183

state: directory setype: 'usr_t' seuser: 'system_u'

After applying the change (using ansible-playbook), the updated definition results in a correctly set SELinux user and SELinux type for this directory. In this case, we added two parameters to the file definition: setype and seuser. The Ansible file module supports the following SELinux-related parameters: • seuser is the SELinux user of the resource. Set this to system_u for system resources, as used in the example. • serole is the SELinux role of the resource. This is generally not used, as role inheritance on the system will generally result in the resource being labeled with the object_r role, which is correct most of the time. • setype is the SELinux type of the resource and is the most commonly used SELinux parameter in file modules. • selevel is the SELinux sensitivity level for the resource. By default, it is set to s0. As we've learned from the example already, you do not need to declare the type if the inherited context is correct.

Loading custom SELinux policies with Ansible Ansible's current release has no support for loading custom SELinux modules. While custom modules are found on Ansible galaxy (the ecosystem where contributors can add more modules), let's see how we would handle distributing a custom policy to the systems under Ansible control and loading the module, but only if it is not loaded yet. While we could start creating custom modules ourselves, let's use a combination of tasks in the existing role to accomplish this. We will try to accomplish the following tasks in sequence: 1. Upload a custom policy called test.cil to the remote system. 2. Check whether this custom policy is already loaded. 3. Load the custom policy, but only if the previous check failed.

184

Configuring SELinux through Infrastructure-as-Code Orchestration

These three tasks are handled through three modules: the copy module, the shell module, and the command module. We will use each of these modules in separate steps: 1. Create the custom policy mentioned earlier in this chapter by placing the test. cil file in the files/ folder of the packt_selinux role. 2. Create a new code block in the main.yml file of the role, with the following content: - name: Upload test.cil file to /usr/share/selinux/custom copy: src: test.cil dest: /usr/share/selinux/custom/test.cil owner: root group: root mode: '0644'

This will ensure that the test.cil file, currently on the master machine, is distributed to the target nodes in the directory we've previously created. 3. Next, we check whether the policy is already loaded. For this, we use the shell module and use the fail or success state later. Hence, we store the return in the test_is_loaded variable, and explicitly tell Ansible to ignore a failure as we use this as a check rather than a state definition: - name: Check if test SELinux module is loaded shell: /usr/sbin/semodule -l | grep -q ^test$ register: test_is_loaded ignore_errors: True

4. The command module loads the policy file, and only if the previous task failed: - name: Load test.cil if not loaded yet command: /usr/sbin/semodule -i /usr/share/selinux/ custom/test.cil when: test_is_loaded is failed

This approach shows how we can use our knowledge of SELinux to define and set states. This method can be used for other SELinux settings as well, for instance, by validating the output of listings (for example, with semanage) before defining or adjusting settings.

Using Ansible for SELinux system administration

Using Ansible's out-of-the-box SELinux support Ansible has quite a few modules available to provide native support for several SELinux-related settings, which we briefly cover here: • The selinux module can be used to set or change the SELinux state (enforcing or permissive) as well as to select the appropriate SELinux policy type (such as targeted): - name: Set SELinux to enforcing mode selinux: policy: targeted state: enforcing

• With the seboolean module, the SELinux booleans can be adjusted at will: - name: Set httpd_builtin_scripting to true seboolean: name: httpd_builtin_scripting state: yes

• The sefcontext module allows us to change SELinux file context definitions: - name: Set the context for /srv/web sefcontext: target: '/srv/web(/.*)?' setype: httpd_sys_content_t state: present

• With selinux_permissive, we can selectively mark certain SELinux policy domains as permissive: - name: Set zoneminder_t as permissive domain selinux_permissive: name: zoneminder_t permissive: true

• The selogin module can be used to map a login to an SELinux user, as with semanage login: - name: Map taylor's login to the unconfined_u user selogin: login: taylor seuser: unconfined_u state: present

185

186

Configuring SELinux through Infrastructure-as-Code Orchestration

• seport can be used to create an SELinux port mapping: - name: Set port 10122 to ssh_port_t seport: ports: 10122 proto: tcp setype: ssh_port_t state: present

Other SELinux settings might be supported through custom modules, but with the method presented earlier, administrators can already start configuring SELinux across all systems in their environment.

Utilizing SaltStack to configure SELinux The second orchestration and automation framework we'll consider is SaltStack, which has commercial backing by the SaltStack company. SaltStack uses a declarative language similar to Ansible and is also written in Python. In this chapter, we will use the open source SaltStack framework, but an enterprise version of SaltStack is available as well, which adds more features on top of the open source one.

How SaltStack works SaltStack, often also described as just Salt, is an automation framework that uses an agent/server model for its integrations. Unlike Ansible, SaltStack generally requires agent installations on the target nodes (called minions) and activation of the minion daemons to enable communications to the master. This communication is encrypted, and the minion authentication uses public-key validation, which needs to be approved on the master to ensure no rogue minions participate in a SaltStack environment. While agent-less installations are possible with SaltStack as well, we will focus on agent-based deployments. In such a configuration, the minions regularly check with the master to see whether any updates need to be applied. But administrators do not need to wait until the minion pulls the latest updates: you can also trigger updates from the master, effectively pushing changes to the nodes. The target state that a minion should be in is written down in a Salt State file, which uses the .sls suffix. These Salt State files can refer to other state files, to allow a modular design and reusability across multiple machines. If we need more elaborate coding, SaltStack supports the creation and distribution of modules, called Salt execution modules. However, unlike Ansible's Galaxy, no community repositories currently exist to find more execution modules.

Utilizing SaltStack to configure SELinux

187

Installing and configuring SaltStack The installation of SaltStack is similar across the different Linux distributions. Let's see how the installation is done on a CentOS machine: 1. We first need to enable the SaltStack repository that contains its software. The project maintains the repository definitions through RPM files that can be installed immediately: # yum install https://repo.saltstack.com/py3/redhat/saltpy3-repo-latest.el8.noarch.rpm

2. Once we have enabled the repository on all systems, install salt-master on the master, and salt-minion on the remote systems: master ~# yum install salt-master remote ~# yum install salt-minion

3. Before we start the daemons on the systems, we first update the minion configuration to point to the master. By default, the minions will attempt to connect to a host with the hostname salt, but this can be easily changed by editing /etc/ salt/minion and setting the right hostname: remote ~# vim /etc/salt/minion master: ppubssa3ed

4. With the minion configured, we can now launch the SaltStack master (salt-master) and minion (salt-minion) daemons: master ~# systemctl start salt-master remote ~# systemctl start salt-minion

5. The minion will connect to the master and present its public key. To list the agents currently connected, use salt-key -L: master ~# salt-key -L Accepted Keys: Denied Keys: Unaccepted Keys: rem1.internal.genfic.local Rejected Keys:

188

Configuring SELinux through Infrastructure-as-Code Orchestration

We need to accept the keys for the remote machines: master ~# salt-key -a rem1.internal.genfic.local The following keys are going to be accepted: Unaccepted Keys: rem1.internal.genfic.local Proceed? [n/Y] y Key for minion rem1.internal.genfic.local accepted.

6. Once we have accepted the key, the master will know and control the minion. Let's see whether we can properly interact with the remote system: master ~# salt '*' service.get_all

This command will list all system services on the minion. The salt command is the main command used to query and interact with the remote minions from the master. If the last command is successfully returning all system services, then SaltStack is correctly configured and ready to manage the remote systems.

Creating and testing our SELinux state with SaltStack Let's create our SELinux state called packt_selinux, and have it applied to the remote minion: 1. We first need to create the top file. This file is the master file for SaltStack, from which the entire environment is configured: master ~# mkdir /srv/salt master ~# vim /srv/salt/top.sls base: '*': - packt_selinux

2. Next, we create the state definition for packt_selinux: master ~# mkdir /srv/salt/packt_selinux master ~# vim /srv/salt/packt_selinux/init.sls /usr/share/selinux/custom/test.cil: file.managed: - source: salt://packt_selinux/test.cil - mode: 644 - user: root - group: root - makedirs: True

Utilizing SaltStack to configure SELinux

189

The init.sls file is the main state file for this packt_selinux state. So, when SaltStack reads the top.sls file, it sees a reference to the packt_selinux state and then searches for the init.sls file inside this state. 3. Place the SELinux test.cil module, as defined earlier on in this chapter, inside /srv/salt/packt_selinux as we refer to it in the state definition. Once placed, we can apply this state to the environment: master ~# salt '*' state.apply

The state.apply subcommand of the salt command is used to apply the state across the environment. Each time we modify our state definition, this command can be used to force an update to the minions. Without this, the minions will (by default) update their state every 60 minutes. These scheduled state updates are called mine updates and are configured on the agents inside /etc/salt/minion.

Assigning SELinux contexts to filesystem resources with SaltStack At the time of writing, support for addressing SELinux types in resources has not yet reached the stable versions of SaltStack. SaltStack, however, supports running commands but only if a certain test has succeeded (or failed). Update the init.sls file and add the following code to it: {%- set path = '/usr/share/selinux/custom/test.cil' %} {%- set context = 'system_u:object_r:usr_t:s0' %} set {{ path }} context: cmd.run: - name: chcon {{ context}} {{ path }} - unless: test $(stat -c %C {{ path }}) == {{ context }}

In this code snippet, we declare two variables (path and context) so that we do not need to iterate the path and context multiple times, and then use these variables in a cmd.run call. The cmd.run approach allows us to easily create custom SELinux support using the commands we've seen earlier on in this book. The unless check contains the test to see whether we need to execute the command or not, allowing us to create idempotent state definitions.

190

Configuring SELinux through Infrastructure-as-Code Orchestration

Loading custom SELinux policies with SaltStack Let's load our custom SELinux module on the remote systems. SaltStack has support for loading SELinux modules through the selinux.module state: load test.cil: selinux.module: - name: test - source: /usr/share/selinux/custom/test.cil - install: True - unless: "semodule -l | grep -q ^test$"

As in the previous section, we need to add an unless statement, as otherwise, SaltStack will attempt to load the SELinux module repeatedly every time the state is applied.

Using SaltStack's out-of-the-box SELinux support SaltStack's native SELinux support is gradually expanding but still has much room for improvement: • With selinux.boolean, the SELinux boolean values can be set on the target machines: httpd_builtin_scription: selinux.boolean: - value: True

• The file contexts, as managed with semanage fcontext, can be defined using the selinux.fcontext_policy_present state: "/srv/web(/.*)?": selinux.fcontext_policy_present: - sel_type: httpd_sys_content_t

• To remove the definition, use the selinux.fcontext_policy_absent definition. • With selinux.mode, we can put the system in enforcing or permissive mode: enforcing: selinux.mode

Automating system management with Puppet

191

• Port mappings are handled using the selinux.port_policy_present state: tcp/10122: selinux.port_policy_present: - sel_type: ssh_port_t

With the cmd.run approach mentioned earlier, we can apply SELinux configuration updates to systems in a repeatable fashion for unsupported settings.

Automating system management with Puppet Puppet is the third automation framework that we will check out. It is the oldest one in our list, with its first release in 2005, and is commonly seen as the baseline against which other automation frameworks are compared. It has commercial backing through the Puppet company, also often referred to as Puppet Labs.

How Puppet works Like SaltStack, Puppet uses an agent/server-based model with public-key authentication of the agents to ensure no rogue agents are active within the environment. The Puppet master has access to the Puppet manifests, which is the declaration of the state that Puppet wants to achieve. These manifests use a specific language inspired by Ruby and can refer to classes provided by modules to ensure reusability across the environment. Puppet modules, hence, are the workhorse within Puppet, and Puppet has a significant community called Puppet Forge that allows you to download and install modules created by the community to more easily manage your environment. Puppet agents will regularly connect to the master, informing the master of the current details of the remote machine. These current details are called facts and can be used by Puppet to dynamically handle changes in the environment. The master then compiles the target state in what it calls a catalog and sends that catalog over to the agent. The agent then applies this catalog and reports the results back.

192

Configuring SELinux through Infrastructure-as-Code Orchestration

Installing and configuring Puppet The Puppet company offers integrated packages for several Linux distributions. The following instructions focus on RPM-compatible distributions, but other platforms have very similar instructions: 1. The Puppet company provides repository definitions through RPM files. After the repositories are established, you can install puppetserver and pdk (on the master) and puppet-agent (on the remote systems) so that the software is readily available to use: # yum install https://yum.puppet.com/puppet6release-el-8.noarch.rpm master ~# yum install puppetserver pdk remote ~# yum install puppet-agent

2. Configure the master to have its certificate properly named. Edit the puppet. conf file inside /etc/puppetlabs/puppet and, within the [master] section, update or add the following settings: master ~# vim /etc/puppetlabs/puppet/puppet.conf certname = ppubssa3ed.internal.genfic.local server = ppubssa3ed.internal.genfic.local environment = production

3. Start the Puppet server so that the clients can start connecting to it: master ~# systemctl start puppetserver

4. On the remote systems, edit the same configuration file, and update or add the following settings in the [main] section: remote ~# vim /etc/puppetlabs/puppet/puppet.conf [main] certname = rem1.internal.genfic.local server = ppubssa3ed.internal.genfic.local environment = production runinterval = 1h

5. Next, start the Puppet agent: remote ~# systemctl start puppet

Automating system management with Puppet

193

6. On the master node, we can now query the pending certificate requests. It should display the requests from the agents we recently started: master ~# /opt/puppetlabs/bin/puppetserver ca list Requested Certificates: rem1.internal.genfic.local (SHA256) ...

7. We can accept this request (sign the certificate) as follows: master ~# /opt/puppetlabs/bin/puppetserver ca sign --certname rem1.internal.genfic.local Successfully signed certificate request for rem1. internal.genfic.local

8. To validate whether the connection works, log in on the remote machine and trigger the agent to apply the (currently empty) catalog: remote ~# /opt/puppetlabs/bin/puppet agent --test

Unlike SaltStack, where we can push a change to the agents, Puppet relies on the agents to frequently poll the server. In the configuration we made earlier, we configured the agent to check every hour. With the puppet agent --test command, we can signal the agent to run the state check immediately.

Creating and testing the SELinux class with Puppet Let's create our packt_selinux class, through which we will configure our remote machine's SELinux settings: 1. Call the Puppet Development Kit (PDK) on the master node inside the /etc/ puppetlabs/code/modules directory: master ~# cd /etc/puppetlabs/code/modules master ~# pdk new module packt_selinux --skip-interview

The result is an empty module with lots of default files and directories. We will be mostly working with the module's manifest file. 2. Inside the packt_selinux/manifests directory, create a new file named init.pp with the following content: class packt_selinux { file { "/usr/share/selinux/custom": ensure => directory, mode => "0755",

194

Configuring SELinux through Infrastructure-as-Code Orchestration } }

3. Next, inside the /etc/puppetlabs/code/environments/production/ manifests location, create a file called site.pp with the following content: node 'rem1.internal.genfic.local' { include packt_selinux }

The site.pp file provides the top-level hierarchy for Puppet to associate its environment with the appropriate definitions. In this example, the node with the hostname rem1.internal.genfic.local is configured through a reference to packt_selinux, the module we created previously. Inside the packt_selinux module, we've created the packt_selinux class, which currently is composed of a single directive to create /usr/share/ selinux/custom. 4. With these definitions in place, have the remote agent update its state: remote ~# puppet agent -t

In product environments, it is common to have this command either scheduled regularly or to run the Puppet agent continuously as a daemon. With the class properly assigned to the node, we can expand our configuration with more SELinux details.

Assigning SELinux contexts to filesystem resources with Puppet Let's augment our current class definition with the following snippet: file { 'selinux_custom_module_test': path => "/usr/share/selinux/custom/test.cil", ensure => file, owner => "root", group => "root", source => "puppet:///modules/packt_selinux/test.cil", require => File["/usr/share/selinux/custom"], seltype => "usr_t", }

Automating system management with Puppet

195

For this block to work properly, we need to place the test.cil SELinux module in the files/ folder inside the packt_selinux module location. This block will have Puppet upload the file to the directory, with the dependency set that the directory must exist. The require statement refers to the previously defined block. We also see that Puppet has out-of-the-box support for SELinux type definitions. The file class has several SELinux-supported parameters that can be used: • seluser defines the SELinux user for the resource. • selrole defines the SELinux role for the resource. • seltype defines the SELinux type for the resource. • selrange defines the SELinux sensitivity range for the resource. • selinux_ignore_defaults tells Puppet to ignore the default SELinux context (as queried from the SELinux policy). Our previous example is thus actually superfluous because Puppet will actively query the SELinux policy to discover what the right resource context is and apply this. With selinux_ignore_defaults set to true, Puppet will not query and adjust the context accordingly, which can be useful when testing out new setups that do not have proper context definitions set.

Loading custom SELinux policies with Puppet Puppet does have support for loading and managing SELinux modules. However, its support is currently restricted to the more traditional SELinux policy modules, and not the CIL powered ones. So, let's create another block in our module definition that loads the test.cil file, but only if no test SELinux module is already loaded: exec { '/usr/sbin/semodule -i /usr/share/selinux/custom/test. cil': require => File['selinux_custom_module_test'], unless => '/usr/sbin/semodule -l | grep -q ^test$', }

This approach allows us to create custom SELinux configuration adjustments if the native Puppet support does not suffice.

196

Configuring SELinux through Infrastructure-as-Code Orchestration

Using Puppet's out-of-the-box SELinux support Puppet has a few SELinux-related classes supported out of the box but has more support through Puppet Forge, an ecosystem of community-contributed modules. One of the modules that we can recommend is the puppet-selinux module, which Puppet (the company) maintains on Puppet Forge (and thus has a higher chance of remaining supported in later versions of Puppet). Installing new modules is quite easy, using the puppet module command: master ~# /opt/puppetlabs/bin/puppet module install puppetselinux

We can then refer to the selinux class (provided through this module) within our manifest: • The selinux class can be directly used to set the enforcing (or permissive) state of the system: class { selinux: mode => 'enforcing', type => 'targeted', }

• The (native) selboolean class can be used to set SELinux booleans: selboolean { 'httpd_builtin_scripting': value => off, }

• SELinux file contexts can be defined using the selinux::fcontext class: selinux::fcontext { '/srv/web(/.*)?': seltype => 'httpd_sys_content_t', }

• Equivalence definitions for the file context are handled by selinux::fcontext::equivalence, like so: selinux::fcontext::equivalence { '/srv/www': ensure => 'present', target => '/srv/web', }

Wielding Chef for system automation

197

• Custom port mappings are handled by selinux::port: selinux::port { 'set_ssh_custom_port': ensure => 'present', seltype => 'ssh_port_t', protocol => 'tcp', port => 10122, }

• Individual SELinux domains can be made permissive using selinux::permissive: selinux::permissive { 'zoneminder_t': ensure => 'present', }

• If standard SELinux modules are present, the use of selmodule allows loading it up. In this case, it will search for the SELinux module named after the block, inside the directory referred to by selmoduledir: selmodule { 'vlock': ensure => 'present', selmoduledir => '/usr/share/selinux/custom', }

While other SELinux-supporting modules might be available on Puppet Forge, be sure to validate whether these modules are mature and sufficiently stable. If their support is uncertain, you might want to pursue the exec route, as used earlier on, in Loading custom SELinux policies with Puppet.

Wielding Chef for system automation The last automation framework we will explore is Chef. Chef is a slightly more hands-on and development-oriented automation framework than the previous ones, but powerful nonetheless. It has commercial backing by the similarly named company Chef.

How Chef works Chef has a slightly more extensive approach to automation and requires slightly more work to get up and running. Once set up, however, it offers a very flexible and programmable environment wherein infrastructure automation can be worked out.

198

Configuring SELinux through Infrastructure-as-Code Orchestration

There are three types of systems in the Chef architecture: • The Chef server acts as the central hub on which the automation code is maintained, and which interacts with the remote systems to apply the changes. • The Chef workstation is an endpoint on which administrators and engineers develop Chef recipes (code) and cookbooks and interact with the Chef server. There can be multiple Chef workstations per Chef environment. • The Chef client is an agent running on the remote systems (nodes) managed by the Chef environment. Developers create automation code in recipes, which are like tasks. Multiple recipes are bundled in a cookbook and uploaded to the Chef server before the recipes can be applied to one or more nodes. Cookbooks can be compared with modules in the previous automation frameworks. The Chef clients and server use public key-based authentication and encryption for their interactions. It is the client that takes the initiative, connecting to the server to download the latest cookbooks and other resources, after which it calculates and applies the latest changes, sending feedback on these changes back to the server.

Installing and configuring Chef A full Chef installation requires a few components to be installed. The Chef workstation and the Chef server need to be installed by the administrator, whereas the Chef agents will be installed by Chef later.

Installing the Chef workstation To install and use Chef, first download the Chef workstation. All Chef software can be downloaded from https://downloads.chef.io. For CentOS, the Chef workstation is available as an RPM, which can be installed using yum. However, unlike common packaged software, the Chef workstation dependencies are not explicitly listed as RPM dependencies, causing the software to be installed without its necessary libraries. At the end of the installation, the RPM file will execute a post-installation script that checks the dependencies and reports on the missing libraries: master ~# yum install chef-workstation-0.17.5-1.el7.x86_64.rpm

The dependencies, currently, require the following CentOS packages to be installed: master ~# yum install libX11-xcb libXcomposite libXcursor libXdamage nss gdk-pixbuf2 gtk3 libXScrnSaver alsa-lib git

Wielding Chef for system automation

199

After the installation, run chef -v (as a regular, non-root user) to verify whether all dependencies are met: master ~$ chef -v

The command should output the versions of the included Chef components.

Installing and configuring the Chef server core The second installation is the Chef server core. This software is again made available as RPM: 1. Install the Chef server core using yum: master ~# yum install chef-server-core-13.2.0-1.el7. x86_64.rpm

After the installation finishes, we need to configure it for our environment. 2. Create a directory named /var/opt/chef. We will use this directory to store the cryptographic keys to authenticate against the Chef server: master ~# mkdir /var/opt/chef

3. Next, configure the Chef server using chef-server-ctl: master ~# chef-server-ctl reconfigure

This will set up the Chef server on the current system. This setup can take a while to complete, but once finished, we can continue with creating a user account inside of Chef. 4. Let's create an account called chefadmin for the user lisa on this system and give it a custom password: master ~# chef-server-ctl user-create chefadmin Lisa McCarthy [email protected] pw4chef --filename /var/opt/chef/chefadmin.pem

5. Create an organization unit inside the Chef configuration, which we associate with the newly created user: master ~# chef-server-ctl org-create ppubssa3ed "Packt Pub SSA 3rd Edition" --association_user chefadmin --filename /var/opt/chef/ppubssa3ed-validator.pem

200

Configuring SELinux through Infrastructure-as-Code Orchestration

With this done, the server administration itself is all done, and we can start creating our development environment.

Preparing the development environment As mentioned earlier on, Chef is somewhat more development-oriented than the previous automation frameworks. The user that will interact with Chef (using the Chef workstation) needs to establish a development environment first: 1. We previously created an account called chefadmin for the user lisa. Now, log in as the user lisa and create a development environment in the user's home directory: master ~$ mkdir chef master ~$ cd chef master ~$ git init

2. We create a Git-enabled environment as the Chef utilities require it. If you have no active Git configuration yet, you might need to add your email and name: master ~$ git config --global user.email "lisa@ ppubssa3ed.internal.genfic.local" master ~$ git config --global user.name "Lisa McCarthy"

3. Next, create the Chef knife configuration as .chef/knife.rb within this environment (so ~/chef/.chef/knife.rb in our example): master ~$ mkdir .chef master ~$ vim .chef/knife.rb current_dir = File.dirname(__FILE__) log_level :info log_location STDOUT node_name "chefadmin" client_key "/var/opt/chef/chefadmin.pem" chef_server_url "https://ppubssa3ed/organizations/ ppubssa3ed" cookbook_path ["#{current_dir}/../cookbooks"]

This configuration references the key used previously as well as the organization we created. If the Chef workstation is a different system than the Chef server, don't forget to copy over the key (chefadmin.pem in our example) and adjust the configuration accordingly.

Wielding Chef for system automation

201

4. Download the certificates that the Chef server uses (these certificates are self-signed certificates) and then check the SSL connection: master ~$ knife ssl fetch master ~$ knife ssl check

5. If the checks are successful, we can commit the changes: master ~$ git add -A master ~$ git commit -m 'Chef configuration baseline'

We are now ready to start our recipe and cookbook development.

Creating the SELinux cookbook The cookbook we are going to develop will contain the various SELinux configuration entries, which are then assigned to the remote node: 1. Let's start by creating a cookbook called packt_selinux: master master master master

~$ ~$ ~$ ~$

mkdir cookbooks cd cookbooks chef generate cookbook packt_selinux cd packt_selinux

This command creates the default files for the cookbook, of which we will handle metadata.rb and recipes/default.rb. The metadata.rb file contains information about the cookbook and, while it is not necessary for our example, it is sensible to edit and update this file immediately. Later, we will adjust this file to include dependency information toward other cookbooks. 2. The recipes/default.rb file contains the actual logic we want to apply to the remote systems. Let's create a definition for the /usr/share/selinux/custom directory: master ~$ vim recipes/default.rb directory '/usr/share/selinux/custom' do owner 'root' group 'root' mode '0755' action :create end

202

Configuring SELinux through Infrastructure-as-Code Orchestration

3. Now upload the cookbook to the Chef server: master ~$ knife cookbook upload packt_selinux

4. We can query the available cookbooks on the Chef server with the list subcommand: master ~$ knife cookbook list packt_selinux 0.1.0

5. With the cookbook available, let's bootstrap the target node. Bootstrapping only needs to occur once, but must be triggered from a Chef authenticated user: master ~$ knife bootstrap rem1 --ssh-user root --nodename rem1

6. This ensures the Chef server knows the remote system. We can query the nodes using knife node list and get more details about a node with the show subcommand: master ~$ knife node show rem1

7. Assign the packt_selinux recipe to the node using the run_list add subcommand: master ~$ knife node run_list add rem1 'recipe[packt_ selinux]'

Adding the recipe to the node list does not automatically trigger the requested update. For this, the remote node's administrator needs to ensure that the chef-client binary executes either regularly (through a cron job or similar) or starts as a daemon. 8. For our purposes, we will trigger the chef-client command on the remote system to download and apply the latest changes: remote ~# chef-client

The output of chef-client should show how it found and applied the changes listed in the recipe. If this command returns successfully, then Chef is ready to manage the remote system using the cookbook we've developed.

Wielding Chef for system automation

203

Assigning SELinux contexts to filesystem resources with Chef Chef has limited native support for SELinux contexts. When instructed to create or modify files on nodes, it will relabel those files according to the present file context definitions on the nodes. We can, however, subscribe to events defined in the recipe, and trigger appropriate actions when they occur. For instance, to explicitly set the context of a directory, we can create something like this: execute 'set_selinux_custom_context' do command '/usr/bin/chcon -t usr_t /usr/share/selinux/custom' action :nothing subscribes :run, 'directory[/usr/share/selinux/custom]', :immediately end

After adding this to the recipes/default.rb file, we first need to upload the updated cookbook to the server: master ~$ knife cookbook upload packt_selinux

Afterward, we can rerun chef-client on the remote node to apply this updated recipe. If the directory was previously already created, the recipe will not change anything as the subscription will not be triggered.

Loading custom SELinux policies with Chef Let's update our recipe to include the logic to load a custom policy. We will use two blocks in our recipe, one to upload the test.cil file to the node, and another one to load it, but only if it was not loaded previously: cookbook_file '/usr/share/selinux/custom/test.cil' do source 'test.cil' owner 'root' group 'root' mode '0755' action :create end bash 'load_test_cil' do code '/usr/sbin/semodule -i /usr/share/selinux/custom/test. cil' not_if '/usr/sbin/semodule -l | grep -q ^test$' only_if { ::File.exists?('/usr/share/selinux/custom/test.

204

Configuring SELinux through Infrastructure-as-Code Orchestration

cil') } end

Put the test.cil file in a folder called files inside the packt_selinux cookbook directory, before uploading the updated cookbook and reapplying the changes using chef-client.

Using Chef's out-of-the-box SELinux support While Chef itself has limited out-of-the-box SELinux support, cookbooks are available online on Chef Supermarket (where the Chef community manages and distributes their custom cookbooks). Chef (the company) maintains the selinux cookbook itself, which allows managing the SELinux state of a system, whereas the selinux_policy cookbook addresses a few other SELinux settings. Let's download and install the selinux and selinux_policy cookbooks: master master master master

~$ ~$ ~$ ~$

knife knife knife knife

supermarket install selinux_policy supermarket install selinux cookbook upload selinux_policy cookbook upload selinux

Next, adjust the metadata.rb file of our own cookbook to include the dependency to this newly added cookbook: depends 'selinux_policy' depends 'selinux'

We can now use some of the predefined recipes to handle SELinux configuration settings: • With selinux_state, we can place the system in an enforcing or permissive state: selinux_state "SELinux enforcing" do action :enforcing end

• The selinux_policy_boolean recipe can configure an SELinux boolean value: selinux_policy_boolean 'httpd_builtin_scripting' do value false end

Summary

205

• With selinux_policy_port, a custom SELinux port mapping can be defined: selinux_policy_port '10122' do protocol 'tcp' secontext 'ssh_port_t' end

• A file context definition can be set using selinux_policy_fcontext: selinux_policy_fcontext '/srv/web(/.*)?' do secontext 'httpd_sys_content_t' end

• An SELinux domain can be put in permissive mode using the selinux_policy_ permissive recipe: selinux_policy_permissive 'zoneminder_t' do end

Don't forget to upload the changed cookbook before calling chef-client on the remote systems.

Summary Automation frameworks such as Ansible, SaltStack, Puppet, and Chef can be easily used to manage SELinux settings on a multitude of systems. While not all frameworks can deal with SELinux settings natively, this is easily mitigated by either using community-provided modules or by creating custom rules that check and update the settings accordingly. In this chapter, we've seen how to accomplish this by installing a custom, CIL-based SELinux policy. We learned that these frameworks all have their specific approaches. Ansible, for instance, does not use any software installations on remote systems and communicates with the target systems using SSH. The other frameworks all use an agent/server model but have their own views on configuring settings (the syntax between Puppet and SaltStack is noticeably different) or design (Chef uses a workstation where developers have their development environment). All these frameworks are easily put in place and configured and can handle most SELinux settings without any problems. All tools have a way of modularizing the definitions so they can be applied easily against a larger number of systems. Now that we know how to apply SELinux settings consistently, let's see what other SELinux controls exist, but now through userspace application-specific support.

206

Configuring SELinux through Infrastructure-as-Code Orchestration

Questions 1. Which of the four tools have native support for setting SELinux contexts on resources? 2. How do these orchestration tools allow reusable customization beyond native support? 3. What are some obvious differences between the listed orchestration tools?

Section 2: SELinux-Aware Platforms

Some applications and platforms have explicit SELinux support to further tighten security controls. In this part, the most common platform SELinux controls are explained. This section comprises the following chapters: • Chapter 7, Configuring Application-Specific SELinux Controls • Chapter 8, SEPostgreSQL – Extending PostgreSQL with SELinux • Chapter 9, Secure Virtualization • Chapter 10, Using Xen Security Modules with FLASK • Chapter 11, Enhancing the Security of Containerized Workloads

7

Configuring Application-Specific SELinux Controls Several Linux services and applications enable additional SELinux controls besides the kernel-enforced SELinux policy. They allow the administrator to further manipulate and enforce policy rules through the application itself—isolating users, reducing data leakage risks, and mitigating the impact of malicious behavior. In this chapter, we will look at several SELinux-aware applications, such as systemd services and how they allow administrators to set up and specify target domains and resource labels. We'll also cover the D-Bus service, which allows SELinux policies to control the service binding and message communication within D-Bus itself. Next, we'll jump to PAM-enabled services that allow users to log in through them. Finally, we'll end the chapter with mod_selinux, an Apache module that allows SELinux-specific tuning of the web server's behavior. This approach shows how applications that do not natively have SELinux support can still be extended to address the administrator's requirements.

210

Configuring Application-Specific SELinux Controls

We will cover the following topics in this chapter: • Tuning systemd services, logging, and device management • Communicating over D-Bus • Configuring PAM services • Using mod_selinux with Apache

Technical requirements Check out the following video to see the Code in Action: https://bit.ly/37jYtze

Tuning systemd services, logging, and device management systemd is a core component of many Linux distributions. Since its birth in 2010, many distributions have gradually adopted systemd as the core init system, responsible for handling services and boot-up operations. Throughout its development phase, systemd added several other components to its portfolio: • D-Bus, which offers a system and session bus service allowing the use of D-Bus for inter-application communication, merged with systemd. • systemd also incorporated udev, which offers a flexible device-node management application. • Login capabilities were added to systemd, enabling fine-grained control over user sessions. • The journald daemon joined the systemd family to provide a new approach to system and service logging, replacing some of the functionality of standard system loggers. • The timerd daemon provides support for the time-based execution of tasks, replacing some of the functionality of standard cron daemons. • Network configurations can be managed by systemd-networkd.

Tuning systemd services, logging, and device management

211

This ongoing approach of absorbing several system services into a single application suite has not gone unnoticed and isn't without controversy. Some distributions even refuse to have systemd as the default init system. The systemd project includes SELinux support for most of its services. Applications such as systemd, which not only include SELinux awareness but also enforce access controls on specific SELinux classes and permissions (rather than relying on the Linux kernel), are called userspace object managers:

Figure 7.1 — The difference between kernel-enforced, standard SELinux, and userspace-managed SELinux

If an application enforces access controls toward certain classes and permissions, then it will also have its own AVC (see Chapter 2, Understanding SELinux Decisions and Logging, for more information about the AVC). Log events resulting from these applications will be identified as USER_AVC events rather than (kernel-managed) AVC events. The systemd application has support for systemd-specific classes, as we will see in the Governing unit operation access section. But before we dive into these specific details, let's first see what systemd is all about and what SELinux support it has.

212

Configuring Application-Specific SELinux Controls

Service support in systemd The main capability of the system daemon that most people know about is its support for system services. Unlike traditional SysV-compatible init systems, systemd does not use scripts to manage services. Instead, it uses a declarative approach for the various services, documenting the wanted state and configuration parameters while using its own logic to ensure that the right set of services start at the right time and in the correct order.

Understanding unit files systemd uses unit files to declare how a service should behave. These unit files use the INI-style syntax, supporting sections and key/value pairs within each file. A service can have multiple unit files that influence the service at large. It is important to remember that different unit files for the same service are all related: • The *.service unit files define how a system service should be launched, what its dependencies are, how systemd should treat sudden failures, and so on. • The *.socket unit files define which socket(s) should be created and which permissions should be assigned to it. systemd uses this for services that can be launched on request rather than directly at boot. • The *.timer unit files define at what time or frequency the service should be launched. Services that do not necessarily run daemonized but need to execute a certain logic at defined intervals can use these timer files to ensure regular runs. These settings are comparable to the more classic yet still widely used crontabs, which we briefly touch upon in PAM services, in the subsection called Cron. Other unit files exist as well, although those have more in common with generic system configurations (such as slice definitions and automount settings) and less with runtime services. System unit files can be placed in one of three locations: • Unit files are installed by default by the system's package manager inside /usr/ lib/systemd/system. • At runtime, updates can be placed inside /run/systemd/system, which will override the unit files in the default location. However, this location is transient and will not persist across reboots. • System administrators can override the configurations in the two locations by placing unit files in /etc/systemd/system. These unit files override previous definitions, so there is no need to remove the unit files from the previous locations.

Tuning systemd services, logging, and device management

213

As an example, check out the default Nginx service unit file, nginx.service, inside / usr/lib/systemd/system: [Unit] Description=The nginx HTTP and reverse proxy server After=network.target remote-fs.target nss-lookup.target [Service] Type=forking PIDFile=/run/nginx.pid ExecStartPre=/usr/bin/rm -f /run/nginx.pid ExecStartPre=/usr/sbin/nginx -t ExecStart=/usr/sbin/nginx ExecReload=/bin/kill -s HUP $MAINPID KillSignal=SIGQUIT TimeoutStopSec=5 KillMode=mixed PrivateTmp=true [Install] WantedBy=multi-user.target

This unit file declares the command to launch Nginx with and informs systemd that the service should be launched after successfully reaching the network, remote-fs, and nss-lookup targets (which is a milestone in the boot process, allowing proper dependency handling). The unit file also declares that it is a dependency of the multi-user target (which is the equivalent of the default run level when using SysV-style init services), which means the service should launch when the system boots.

Setting the SELinux context for a service When systemd launches a service, it executes the command defined through the ExecStart= configuration entry in the service unit file. By default, a standard domain transition will occur as defined through the SELinux policy. Package developers and system administrators can, however, update the service unit files to have the service launched in an explicitly mentioned SELinux domain. To accomplish this, the [Service] section of the unit file can be extended with the SELinuxContext= configuration entry.

214

Configuring Application-Specific SELinux Controls

For instance, to ensure that Nginx launches with the httpd_t:s0:c0.c128 context, you'd use this: [Service] Type=forking PIDFile=/run/nginx.pid ExecStartPre=/usr/bin/rm -f /run/nginx.pid ExecStartPre=/usr/sbin/nginx -t ExecStart=/usr/sbin/nginx ExecReload=/bin/kill -s HUP $MAINPID SELinuxContext=system_u:system_r:httpd_t:s0:c0.c128 KillSignal=SIGQUIT TimeoutStopSec=5 KillMode=mixed PrivateTmp=true

Of course, it is also possible to use this to have a service running with a different context, which can be useful when developing custom policies for daemons. However, keep in mind that the SELinux policy rules still apply: you cannot ask systemd to launch Nginx, for instance, with the dnsmasq_t domain without updating the SELinux policy so that httpd_exec_t (the entry point for the httpd_t domain) is also made an entry point for the dnsmasq_t domain. When you request systemd to explicitly use an SELinux context for a service, systemd will attempt to use this context for all execution-related tasks: ExecStartPre, ExecStart, ExecStartPost, ExecStopPre, ExecStop, ExecStopPost, and ExecReload. As these tasks often are not labeled with the right entry point label, these commands can fail. In that case, prefix the commands with + so that the SELinux context definition does not apply to them: [Service] Type=forking PIDFile=/run/nginx.pid ExecStartPre=+/usr/bin/rm -f /run/nginx.pid ExecStartPre=/usr/sbin/nginx -t ExecStart=/usr/sbin/nginx ExecReload=/bin/kill -s HUP $MAINPID SELinuxContext=system_u:system_r:httpd_t:s0:c0.c128 KillSignal=SIGQUIT TimeoutStopSec=5 KillMode=mixed PrivateTmp=true

Tuning systemd services, logging, and device management

215

While developing and changing unit files, the changed settings might not always be immediately applied to the system. Running systemctl daemon-reload after modifying unit files will ensure that the latest changes on the system are read by systemd.

Using transient services systemd can also be used to launch applications as if they are services and have them under systemd's control. Such applications are called transient services as they lack the unit files that generally declare how systemd should behave. Transient services are launched through the systemd-run application. To show this, let's create a simple Python script (one that calculates Pi up to 10,000 digits): from decimal import Decimal, getcontext getcontext().prec=10000 with open('/tmp/pi.out', 'w') as f: print(sum(1/Decimal(16)**k * ( Decimal(4)/(8*k+1)Decimal(2)/(8*k+4)Decimal(1)/(8*k+5)Decimal(1)/(8*k+6)) for k in range(10000)), file=f)

As this takes some time, we can opt to run this Python script under systemd's control: # systemd-run python3.6 /tmp/pi.py Running as unit: run-rf9ce45c...f343.service

As transient services do not have unit files to manage, changing the SELinux context must be accomplished through the command line as well. Of course, this is only needed if the standard domain transitions defined in the policy do not result in the wanted behavior: # systemd-run -p SELinuxContext=guest_u:guest_r:guest_t:s0 python3.6 /tmp/pi.py

The systemd-run application supports this through the --property (or -p) option, through which unit file properties can be added. In the previous example, we use this option to run the script in the guest_t domain using the SELinuxContext property, similar to how we would define this in the unit file itself.

Requiring SELinux for a service Some services should only run when SELinux is enabled or disabled. With systemd, this can be defined through its conditional parameters.

216

Configuring Application-Specific SELinux Controls

A service unit file can contain several conditions that need to be valid before systemd will consider executing the service. These conditionals can point to the system type (virtualized or not), kernel command-line parameters, files that do or don't exist, and so on. The one we are interested in is ConditionSecurity, which represents the state of the given security system—in our case, SELinux. For instance, look at the selinux-autorelabel.service unit file inside /usr/ lib/systemd/system: [Unit] Description=Relabel all filesystems DefaultDependencies=no Conflicts=shutdown.target After=sysinit.target Before=shutdown.target ConditionSecurity=selinux [Service] ExecStart=/usr/libexec/selinux/selinux-autorelabel Type=oneshot TimeoutSec=0 RemainAfterExit=yes StandardOutput=journal+console

Similarly, the Linux distribution provides the selinux-autorelabel-mark. service file. This service ensures that, if SELinux is not active when the system boots (and no /.autorelabel file exists yet), then systemd will create an empty /.autorelabel file. This file ensures that, when the system reboots with SELinux support, the relabeling operation occurs.

Relabeling files during service startup One of the actions that many services require is the preparation of service-specific runtime directories, such as /run/httpd for the Apache service. systemd supports this through tmpfiles.d. We have briefly covered tmpfiles in Chapter 4, Using File Contexts and Process Domains. Within tmpfiles, we can define the files and locations requested to be provided or updated immediately (at boot time) when these are not placed in the (persisted) filesystem. For instance, the package that provides the Apache daemon installs the following definition as /usr/lib/tmpfiles.d/httpd.conf on the system: d /run/httpd 710 root apache d /run/httpd/htcacheclean 700 apache

Tuning systemd services, logging, and device management

217

Like the systemd unit files, the files that contain these settings should be declared in one of the following three locations. Each location overrides the settings of the previous one: • The default, package-provided location is /usr/lib/tmpfiles.d. • Runtime declarations can be placed in /run/tmpfiles.d. • Local system administrator-provided declarations are placed in /etc/ tmpfiles.d. These definitions can get much more specific than just directory creation. Through the tmpfiles.d application, definitions can be set to create files, empty directories upfront, create sub-volumes, manage special files such as symbolic links or block devices, set extended attributes, and more. One of its features is to set the file mode and ownership, and restore the SELinux context on a file (z) or recursively against a directory (Z). This can be used to change contexts on files that have a proper context definition in the policy, but whose context is not properly assigned. For instance, look at the definitions in the selinux-policy.conf file inside /usr/ lib/tmpfiles.d: z Z z w

/sys/devices/system/cpu/online - - /sys/class/net - - /sys/kernel/uevent_helper - - /sys/fs/selinux/checkreqprot - - - - 0

We need to relabel files inside /sys because this location is labeled with sysfs_t by default and changing the context at runtime does not preserve its status across reboots. Yet some of its files should have a different label – the /sys/devices/system/cpu/ online file, for instance, requires the cpu_online_t label: # matchpathcon /sys/devices/system/cpu/online /sys/devices/system/cpu/online system_u:object_r:cpu_ online_t:s0

The definition ensures that this (pseudo) file is relabeled at boot so that all other processes that rely on the file labeled with cpu_online_t can happily continue working. The other arguments to the definition are explicitly marked with a dash in the previous example, meaning that no other parameters need to be configured. They can be used to set the mode, User Identifier (UID), Group Identifier (GID), age, and argument related to the rule.

218

Configuring Application-Specific SELinux Controls

An example configuration that uses some of these other parameters with the z or Z state is the systemd.conf file: # z Z z z z -

grep ^[zZ] /usr/lib/tmpfiles.d/systemd.conf /run/log/journal 2755 root systemd-journal - /run/log/journal/%m ~2750 root systemd-journal - /var/log/journal 2755 root systemd-journal - /var/log/journal/%m 2755 root systemd-journal - /var/log/journal/%m/system.journal 0640 root systemd-journal -

For more information about the definition format, see man tmpfiles.d.

Using socket-based activation The system daemon also supports socket-based activation. When configured, systemd will create the socket on which the daemon usually listens and will have the daemon launched when the socket is first used. This allows systems to boot quickly (as many daemons do not need to be launched immediately) while still ensuring that all required sockets are available. When a client only writes information to the socket (such as with the /dev/log socket), the client does not even need to wait for the daemon to be activated. The data is stored in a buffer until the daemon can read it. Only when the buffer is full will the operation block until the daemon flushes the buffer. Take a look at the systemd-journald.socket unit file, available inside /usr/lib/ systemd/system: [Unit] Description=Journal socket Documentation=man:systemd-journal.service(8) man:journald. conf(8) DefaultDependencies=no Before=sockets.target IgnoreOnIsolate=yes [Socket] ListenStream=/run/systemd/journal/stdout ListenDatagram=/run/systemd/journal/socket SocketMode=0666 PassCredentials=yes PassSecurity=yes ReceiveBuffer=8M Service=systemd-journald.service

Tuning systemd services, logging, and device management

219

When a client uses one of the mentioned sockets, then systemd will launch the systemd-journald.service unit to accommodate the client interaction. As long as these sockets are not used, the service will not be started. Inside the [Socket] section, an SELinux-specific entry can be defined: SELinuxContextFromNet=true. When a unit file has this entry set, systemd will obtain the MLS/MCS information from the client context (the application connecting to the socket) and append this to the context of the service. This sensitivity inheritance can be used to prevent any information leakage from taking place when communication is happening through sockets.

Governing unit operation access Until now, we've looked at configuration settings related to systemd's SELinux support. systemd also uses SELinux to control access to services defined through unit files. When a user wants to perform an operation against a unit (such as starting a service or checking the state of a running service), systemd queries the SELinux policy to see whether it will allow this operation. The systemd daemon uses the service class to validate the permissions of the client's domain toward the requested operation. For instance, to validate whether a user context, sysadm_t, can view the status of the service associated with the sshd.service unit file, it checks the context of this file (being sshd_unit_file_t) and then validates whether the status permission is granted: # sesearch -s sysadm_t -t sshd_unit_file_t -c service -p status -A

Other supported permissions are disable, enable, reload, start, and stop. When a permission is not granted, a USER_AVC denial message will be visible in the audit logs (rather than an AVC message) as the message is not generated by the Linux kernel, but by systemd. So, while the rules themselves are part of the SELinux policy, it is systemd that enforces the access. systemd, or the client through which systemd is queried, might also provide additional error messages to reflect that the SELinux policy prevents the action. For instance, if we attempt to query systemd over D-Bus (which we cover in the D-Bus communication section) from an unprivileged user domain, then we get the following error: Error: GDBus.Error:org.freedesktop.DBus.Error.AccessDenied: SELinux policy denies access

220

Configuring Application-Specific SELinux Controls

To facilitate troubleshooting any systemd-triggered failures, systemd also has an extensive logging component, called systemd-journald, which we'll cover next.

Logging with systemd systemd is not only responsible for service management: it takes up several other tasks as well. One of these tasks is log management, traditionally implemented through a system logger. While systemd still supports running with a traditional system logger, it now suggests the use of systemd-journald. One of the advantages of the journal daemon is that it is not limited to textual, single-line log messages. Daemons can now use binaries as well as multiline messages as part of its logging capabilities. The journal daemon also registers information about the sending process alongside the log messages themselves. This additional information contains ownership data (the process owner) including the SELinux context of the sending process.

Retrieving SELinux-related information The traditional approach to receive SELinux-related information (excluding the audit events we tackled before) is to grep through the log information. With the journal daemon, we can accomplish this as follows: # journalctl -b | grep -i selinux

The -b option passed on to the journal control application informs the journal daemon that we are only interested in the log messages that originated for a specific boot.

Querying logs given an SELinux context A unique feature of the journal daemon is to use the information associated with the log messages as part of the query to be launched against the journal database. For instance, we can ask the journal daemon to only show those messages that originated from a daemon or application running in the udev_t context: # journalctl _SELINUX_CONTEXT=system_u:system_r:init_t:s0

The available contexts can be retrieved through the Bash completion support on the system. After writing _SELINUX_CONTEXT=, press Tab twice to see the possible values.

Tuning systemd services, logging, and device management

221

Using setroubleshoot integration with journal The SELinux troubleshoot daemon is also integrated with systemd-journald. Any alert that comes up from setroubleshootd is also available through the journal daemon. This helps administrators as they will quickly find out about SELinux denials when investigating problems. For instance, when the Nginx web server is not working properly and this is due to an SELinux policy, a quick investigation of the status of the service will reveal that the SELinux policy is preventing some actions: # systemctl status nginx

To get more information about the message, use journalctl: # journalctl -xe

As you can see, systemd-journald has captured environment information related to the service, which can provide much-needed guidance on resolving potential problems. A third systemd service that has SELinux configuration possibilities is the device daemon.

Handling device files Linux has a long history of device managers. Initially, administrators needed to make sure that the device nodes were already present on the filesystem (/dev was part of the persisted filesystem). Gradually, Linux adopted more dynamic approaches for device management. Nowadays, device files are managed through a combination of a pseudo filesystem (devtmpfs) and a userspace device manager called udev. This device manager is merged in systemd as well, becoming systemd-udevd. The device manager listens on a kernel socket for kernel events. These events inform the device manager about detected or plugged-in devices (or the removal of such devices) and allow the device manager to take appropriate action. For udev, these actions are defined in udev rules.

Using udev rules Configuring the udev subsystem is mainly done through udev rules. These rules are one-liners that contain a matching part and an action part.

222

Configuring Application-Specific SELinux Controls

The matching part contains validations, executed against the event(s) that udev receives from the Linux kernel. This validation uses key/value pairs obtained from the event, and includes the following possible keys: • Kernel-provided device name (KERNEL) • Device subsystem (SUBSYSTEM) • Kernel driver (DRIVER) • Specific attributes (ATTR) • Active environment variables (ENV) • The action type to inform if the device is detected or removed (ACTION) While more match keys are possible, the preceding list is most commonly used. The Linux kernel will also inform the device manager about the device hierarchy. This allows rules to be defined based on, for instance, the USB controller through which a USB device is plugged in. Alongside the information for the device itself, the kernel will also provide hierarchically related information through similar key/value pairs. These pairs, however, use a key definition in plural form: SUBSYSTEMS instead of SUBSYSTEM, DRIVERS instead of DRIVER, and so on. For instance, to match a USB webcam with vendor ID 05a9 and product ID 4519, the match-related pairs could look like this: KERNEL=="video[0-9]*", SUBSYSTEM=="video4linux", SUBSYSTEMS=="usb", ATTR{idVendor}=="05a9", ATTR{idProduct}=="4519"

The second part of a udev rule is the action to take. The most common action is to create a symbolic link to the created device file, ensuring that applications can always reach the same device through the same symbolic link, even when the device from the kernel point of view has a different name. We can, for instance, extend the preceding example with SYMLINK+="webcam1" to have /dev/webcam1 point to this newly detected device. The udev application supports many more actions than just defining symbolic links, of course. It can associate ownership (OWNER) or group membership (GROUP) on the device, controlling who can access the devices. udev can also set environment variables (ENV) and even run a command (RUN) when the matched device is plugged in or detached from the system. To make sure the command is only executed when the device is added, we need to add an ACTION setting such as ACTION=="add".

Tuning systemd services, logging, and device management

223

Important note udev can interpret ENV as both a matching key as well as an action key. The difference is the operation performed (a single equals sign = or a double ==). ENV{envvar}=="value" is a match operation (checking whether the variable matches the given value), whereas ENV{envvar}="value" is an action (setting the variable to value).

udev rules are provided by default through the /usr/lib/udev/rules.d location. Distributions and applications/drivers will store their default rules in this location. Additional rules or rule overrides can be placed in /etc/udev/rules.d. It's important to remember that udev will continue processing rules even when it has already encountered a matching rule. This can be changed on a per-rule basis through the OPTIONS action, as with OPTIONS+="last_rule", which informs udev that it can stop processing further rules for this event.

Setting an SELinux label on a device node One of the actions that udev supports is to assign an SELinux context on the device node. We can do this using the SECLABEL{selinux} action: KERNEL=="fd0", ..., SECLABEL{selinux}="system_u:object_r:my_ device_t:s0"

Note that this action only sets the context on the device node. If the rule also sets a symbolic link, then the symbolic link itself will inherit the default device_t context. Placing an SELinux label on a device node is often done together with the other security-related permissions, so the rule often receives additional actions such as setting the target owner (OWNER), group (GROUP), and permission set (MODE). After all, SELinux security controls only apply after the regular, discretionary access control checks have passed, so don't forget to make sure your users have access to the device nodes outside of the SELinux controls as well. All the settings we've seen so far are about systemd service management and system support. Another component within the systemd ecosystem is D-Bus, which is less about system management and more about facilitating communication and interaction between different applications over a programmable communication bus.

224

Configuring Application-Specific SELinux Controls

Communicating over D-Bus The D-Bus daemon provides an inter-process communication channel between applications. Unlike traditional IPC methods, D-Bus is a higher-level communication channel that offers more than simple signaling or memory sharing. Applications that want to chat over D-Bus link with one of the many D-Bus-compatible libraries, such as those provided by the libdbus, sd-bus (part of systemd), GDBus, and QtDBus applications. The D-Bus daemon is part of the systemd application suite.

Understanding D-Bus Linux generally supports two D-Bus types – system-wide and session-specific D-Bus instances: • The system-wide D-Bus is the main instance used for system communication. Many services or daemons will associate themselves with the system D-Bus to allow others to communicate with them through D-Bus. • The session-specific D-Bus is an instance running for each logged-in user. It is commonly used by graphical applications to communicate with each other within a user session. Both D-Bus instances are provided through the dbus-daemon application. The system-wide D-Bus will run with the --system option, whereas a session-specific instance will run with the --session option. Applications register themselves against D-Bus through a namespace. Conventionally, this namespace uses the domain name of the project. For instance, systemd declares the org.freedesktop.systemd1 namespace, whereas D-Bus is at org.freedesktop. DBus. The currently associated applications can be queried using Python easily: # python3.6 >>> import dbus >>> for service in dbus.SystemBus().list_names(): ... print(service) org.freedesktop.DBus org.freedesktop.login1 org.freedesktop.systemd1 org.freedesktop.PolicyKit1 com.redhat.tuned :1.10 :1.11

Communicating over D-Bus

225

org.freedesktop.NetworkManager ...

Each application then provides objects on the bus that can be reached by other objects (other applications)—of course, assuming they have the privileges to do so. These objects are represented through a path-like syntax and generally also use the domain of the project as a prefix. For instance, to list the objects currently associated with org.freedesktop. systemd1, we can use the gdbus command. To facilitate its use, we first enable auto-completion support, after which we can use the Tab key to easily add the appropriate values: # source /usr/share/bash-completion/completions/gdbus # gdbus call --system --dest # gdbus call --system --dest org.freedesktop.systemd1 --objectpath /org/freedesktop/systemd1 Display all 220 possibilities? (y or no) /org/freedesktop/systemd1 /org/freedesktop/systemd1/job /org/freedesktop/systemd1/unit ...

Applications can trigger methods on these objects, or send messages to the applications bound to these objects through these methods. For instance, to get the state of the sshd.service unit through D-Bus, we invoke the org.freedesktop.systemd1.Manager.GetUnitFileState method on the org.freedesktop.systemd1 object reachable through the /org/freedesktop/ systemd1 path, and with the sshd.service argument, like this: # gdbus call --system \ --dest org.freedesktop.systemd1 \ --object-path /org/freedesktop/systemd1 \ --method org.freedesktop.systemd1.Manager.GetUnitFileState \ sshd.service ('enabled',)

These calls can also be controlled through the SELinux policy, as we will learn next.

226

Configuring Application-Specific SELinux Controls

Controlling service acquisition with SELinux The D-Bus application, like systemd, will query the SELinux policy to verify whether to allow an operation. Again, it is the D-Bus application itself that enforces the policy and not a Linux kernel subsystem. The first control that administrators can enable within D-Bus is to ensure that only well-established domains can acquire a specified object within D-Bus. Without this control, malicious code could register itself as org.freedesktop.login1, for instance, and act as a system daemon on the bus. Other applications might mistakenly send out sensitive information to the application. Applications store this policy information in files hosted in /usr/share/dbus-1/ system.d. The login service, for instance (stored as org.freedesktop.login1. conf) has the following policy snippet installed:





...

As the login daemon runs in the systemd_logind_t domain, we could enhance this configuration as follows:



...

Communicating over D-Bus

227

With this enhancement in place, D-Bus will check whether the application (which we presume is running in the systemd_logind_t context) has the acquire_svc permission (of the dbus class) against the systemd_logind_t context. By default, the SELinux policy does not have this permission, and as such, the registration fails: # systemctl restart dbus-org.freedesktop.login1 Job for systemd-logind.service failed because a timeout was exceeded. See "systemctl status systemd-logind.service" and "journalctl -xe" for details. # ausearch -m user_avc -ts recent

When we add the following SELinux policy rule, the registration of systemd-logind will succeed, as expected: (allow systemd_logind_t systemd_logind_t (dbus (acquire_svc)))

Load this policy (say test.cil) and try the restart operation again: # semodule -i test.cil # systemctl restart dbus-org.freedesktop.login1

By limiting which domains can obtain a given service, we ensure that only trusted applications are used. Non-trusted applications will generally not run within the domain of that application (end users, for instance, cannot trigger a transition to such a domain) even if they receive root privileges (which is another check that D-Bus does for the login service, as shown in the first busconfig snippet). Administrators can enhance this D-Bus configuration without having to alter the existing configuration files. For instance, the previously mentioned SELinux-governing busconfig snippet could very well be saved as a different file.

Governing message flows A second control that D-Bus validates is which applications can communicate with each other. This is not configurable through the service configurations but is a pure SELinux policy control. Whenever a source application is calling a method of a target application, D-Bus validates the send_msg permission between the two domains associated with the source and target applications.

228

Configuring Application-Specific SELinux Controls

For instance, communication over D-Bus between a user domain (sysadm_t) and service domain (systemd_logind_t) will check the following permissions: allow sysadm_t systemd_logind_t : dbus send_msg; allow systemd_logind_t sysadm_t : dbus send_msg;

If these permissions are not granted, then D-Bus will not allow the communication to happen. If at any point, the application context cannot be obtained, then the bus daemon context will be used. Failures will be logged as USER_AVC entries in the audit log. If the communication should be allowed, we can create a simple SELinux policy file to address this like so: (allow sysadm_t systemd_logind_t (dbus (send_msg))) (allow systemd_logind_t sysadm_t (dbus (send_msg)))

Store these rules in a file with the suffix .cil (say, local_logind_systemd.cil), and load it with semodule: # semodule -i local_logind_systemd.cil

Let's consider a few other applications that have SELinux support, not necessarily built-in, but through the SELinux policy and PAM integration within the system.

Configuring PAM services systemd and D-Bus are SELinux-aware applications, with explicit SELinux support built in. Several other services exist on a Linux system that play nicely together with SELinux yet are not SELinux-aware themselves. Many of these services have an affinity with SELinux through their PAM integration. We covered PAM integration in Chapter 3, Managing User Logins. In this section, we'll cover three example services using PAM, and how SELinux can be further fine-tuned to support these services.

Cockpit Cockpit is a simple, browser-based management application that allows administrators to easily see system resources (monitoring) as well as to interact with the system. It also allows users to log into the system through the browser. It is this browser-based terminal that we want to configure: by tuning the target SELinux roles for the SELinux users, we can selectively put users in a specific role. This effectively defines what the users can accomplish through this browser-based session.

Configuring PAM services

229

Installing Cockpit The Cockpit application is readily available in the CentOS repository, so installing it is a breeze: # yum install cockpit

While the application does not need additional configuration, if you do need tweaks, you will need to create the configuration file, /etc/cockpit/cockpit.conf, yourself as the application does not create a default configuration file. Within this configuration file, you can configure the TLS settings, or disable encrypted communication generally. Let's disable the encrypted communication for this demonstration run (but if you intend to use Cockpit in production, you should not only keep encryption on but also ensure that only trusted hosts are connecting, possibly even requiring client certificate authentication using the ClientCertAuthentication directive): [WebService] AllowUnencrypted=true

With this set, we can continue with configuring SELinux for Cockpit.

Restricting user logins Through these instructions, we will add the more restricted user_r role to the staff_u SELinux user, and then ensure that all logins mapped to the staff_u SELinux user are logged in using the user_r role when they log in through Cockpit. If they log in through other services, they will continue using the default staff_r role. Note The use of the user_r role rather than the (even more restricted) guest_r role is to allow the Cockpit application to function properly. The application will run a service under the user's privileges, which are not sufficient for Cockpit if we use the guest_t user domain.

Let's first add the user_r role so that we can put the users in the correct context later: # semanage user -m -R "staff_r sysadm_r system_r user_r" staff_u

230

Configuring Application-Specific SELinux Controls

Next, we want to update the SELinux configuration so that any Cockpit login by staff_u mapped users is going to use the user_r role. The Cockpit application has logins done through a service running in the cockpit_session_t context, which we find out by checking the context of the process first, and then logging in on Cockpit and checking the context of the processes again. There, we notice that a new process (cockpit-session) runs with the cockpit_session_t context: # ps -eZ | grep cockpit system_u:system_r:cockpit_ws_t:s0 ... cockpit-ws system_u:system_r:cockpit_session_t:s0 ... cockpit-session localhost

With this information now available, we can edit the /etc/selinux/targeted/ contexts/users/staff_u file as follows: system_r:local_login_t:s0 staff_r:staff_t:s0 sysadm_r:sysadm_t:s0 system_r:remote_login_t:s0 staff_r:staff_t:s0 system_r:sshd_t:s0 staff_r:staff_t:s0 sysadm_r:sysadm_t:s0 system_r:cockpit_session_t:s0 user_r:user_t:s0 system_r:crond_t:s0 staff_r:staff_t:s0 staff_r:cronjob_t:s0

By adjusting the order of the roles listed for the cockpit_session_t context (or limiting them to only the user_r role), we ensure that users allowed to run with the user_r role (like the staff_u user we configured earlier on) do so through the user_r role. As this role is more restricted than the default staff_t user domain, logins through Cockpit are thus more isolated. This approach can be used for all PAM-enabled services, as this solely relies on the pam_ selinux.so call in the service PAM configuration. For some services, the SELinux policy administrators add in a few more tweaks to use, such as with cron and SSH, which we'll discuss next.

Cron Cron services on a system allow you to run tasks or commands on predefined schedules. Some cron applications are explicitly made SELinux-aware (such as fcron), allowing them to compute the target context a job should run in. Even cron systems that do not have any specific SELinux logic built in can be fine-tuned.

Configuring PAM services

231

Switching between user-specific and generic contexts A common setup supported through the SELinux policy is to toggle whether user tasks run in the user's default context (such as staff_t for staff users) or in a default, restricted cron context (cronjob_t). Both approaches have their pros and cons. When we configure the system to have user jobs run in the user's default context, then users know what the privileges are of their jobs. A guest user has guest privileges, a staff user has staff privileges, and so forth. This is the most common configuration, and the default cron system on CentOS uses the context of the file containing the user's tasks (located in /var/spool/cron) to deduce the target runtime context. By running user jobs in a more restricted context such as cronjob_t, all users' cron jobs run with the same privileges, and the administrator can easily fine-tune the privileges for all user jobs. This also allows the administrator to grant specific privileges for cron jobs while keeping the user contexts free of these rights. Let's have a simple task executed every minute, namely a 59-second sleep. As a regular user, create a file (let's say lisa.cron) with the following content: * * * * * sleep 59

This file uses the common cron syntax, where the following applies: 1. The first field covers the minute. 2. The second field covers the hour. 3. The third field covers the day of the month. 4. The fourth field covers the month. 5. The fifth field covers the day of the week. 6. The rest of the line is the command to execute. The fields can use expressions to facilitate time definitions. For instance, to run every 15 minutes, you can use */15 in the first field. If you want to run only at 8 o'clock and 18 o'clock, you can use the 8,18 value in the second field. Another example is if you only want to run on workdays, for which you can use 1-5 in the fifth field (in cron, Sunday holds both 0 and 7 as valid values).

232

Configuring Application-Specific SELinux Controls

By loading it with the crontab command, the file is checked for errors and, if error-free, is securely placed inside /var/spool/cron (the crontab command is a setuid command that is able to modify /var/spool/cron even though this location is inaccessible by regular users): $ crontab ./lisa.cron

From here, the cron daemon will pick up this file, and 1 minute later we will see the command active in the background: $ ps -efZ | grep sleep staff_u:staff_r:staff_t:s0 ...

sleep 59

As seen from the output, the command is running in the staff_t context. To change this to the cronjob_t type, rather than editing the SELinux context definition file as we did with the Cockpit application, use the cron_userdomain_transition SELinux boolean: # setsebool cron_userdomain_transition off

This boolean changes the active SELinux policy behavior so that any user task executed from the cron system executes within the cronjob_t domain. You might need to reset the crontab definition (this depends on the cron system used), but afterward, we will see the job running in the cronjob_t domain: $ ps -efZ | grep sleep staff_u:staff_r:cronjob_t:s0 ...

sleep 59

The use of SELinux booleans to allow administrators to differentiate system behavior as needed is commonly used. For the SSH daemon, SELinux policy administrators have defined something similar.

OpenSSH The OpenSSH daemon is the most common secure shell daemon around. It allows users to remotely access systems through a terminal, as well as to securely transfer files, tunnel application communications, and more. When logging in through SSH, the PAM controls apply, but the SELinux policy also has specific SSH controls embedded and controllable through SELinux booleans.

Configuring PAM services

233

Directly logging in as sysadm_t The first change to assess is to allow directly logging in using the sysadm_r role. Users mapped to the staff_u SELinux user by default log in using the (more restricted) staff_r role, and then need to explicitly switch roles to obtain the more privileged sysadm_r role. The first change we need to make is to edit the /etc/selinux/targeted/ contexts/users/staff_u file and adjust the order of the roles listed for the sshd_t context: system_r:local_login_t:s0 staff_r:staff_t:s0 sysadm_r:sysadm_t:s0 system_r:remote_login_t:s0 staff_r:staff_t:s0 system_r:sshd_t:s0 sysadm_r:sysadm_t:s0 staff_r:staff_t:s0 system_r:cockpit_session_t:s0 user_r:user_t:s0 system_r:crond_t:s0 staff_r:staff_t:s0 staff_r:cronjob_t:s0

However, this is not enough. The SELinux policy administrators have disabled direct logins through SSH to the sysadm_r role, forcing users to explicitly change roles (and thus reauthenticate). This approach is because SSH is often a publicly reachable and not otherwise easily controllable service (unlike services such as web servers, which can have reverse proxies and web application firewalls in front). Change the SELinux ssh_sysadm_login boolean to true to enable the wanted behavior: # setsebool ssh_sysadm_login true

This boolean changes the SELinux policy behavior to allow logins to the sysadm_r role from the SSH daemon.

Chrooting Linux users Another feature that SSH supports is forcing logins from selected users to be chrooted. A chroot (which is a portmanteau of change root) is an isolation method for processes, where the process no longer sees the entire filesystem but only a part of it.

234

Configuring Application-Specific SELinux Controls Informational note Now, chroot environments are an easy way to isolate processes, but a chroot itself is still governed through Linux's discretionary access controls, and escaping chroot environments is not impossible. Using SELinux to further confine the process is recommended but is not in the scope of this section. For that, we refer to Chapter 14, Dealing with New Applications.

Before we configure SSH to chroot some users, we need to create a properly functioning environment: once we change the root for a process, all commands and libraries that the process wants to read or execute need to be available within this chroot environment. Let's first create a chroot environment. A nice utility that assists in creating the right folder structure and files is Jailkit. Jailkit is not available by default through the regular repositories but can be easily installed and only requires a working compiler and Python environment. We start off by installing the necessary dependencies: # yum install gcc python36-devel

Next, we download the Jailkit source code and build it. As CentOS does not have a linked Python binary by default (as it requires the use of python3 as the runtime), we need to tell the build scripts how to address Python. We do this by declaring the PYTHONINTERPRETER environment variable: # # # # # # #

wget https://olivier.sessink.nl/jailkit/jailkit-2.21.tar.bz2 tar xvf jailkit-2.21.tar.bz2 cd jailkit-2.21 export PYTHONINTERPRETER=/usr/bin/python3 ./configure make make install

Once the installation is complete, you might need to remove a duplicate includesections call within the Jailkit configuration file (the jk_init command, which we will use next, will inform you about it if you don't). The openvpn section in / etc/jailkit/jk_init.ini should look like this: [openvpn] comment = jail for the openvpn daemon paths = /usr/sbin/openvpn users = root,nobody

Configuring PAM services

235

groups = root,nobody devices = /dev/urandom, /dev/random, /dev/net/tun includesections = netbasics, uidbasics need_logsocket = 1

With the configuration updated, we can now create the chroot environment. Let's create the /srv/chroot directory and then populate it with the necessary files, directories, device nodes, and more with the jk_init command: # mkdir /srv/chroot # jk_init -v -j /srv/chroot extshellplusnet

We want to make sure that the SELinux contexts for the resources inside this location are equivalent to the root location, so let's create a file context equivalency definition: # semanage fcontext -a -e / /srv/chroot # restorecon -RvF /srv/chroot

With the chroot environment set, we can now update the SSH configuration to chroot a user: Match User lisa X11Forwarding no AllowTcpForwarding no ChrootDirectory /srv/chroot

While not applicable to all systems (as it depends on the distribution), we might need to tell the SELinux policy that the user domains for the users can chroot. This privilege (sys_chroot) is often not enabled by default for user domains: # setsebool selinuxuser_use_ssh_chroot true

With this set, restart the SSH daemon and see whether the chroot is successful: # systemctl restart ssh

Chroot environments are not only sensible for SSH access; other daemons might support chroot environments to further protect the resources on the system. In the past, chroot support was a common way to further harden the system. Namespace and resource isolation support has, however, largely surpassed the need for chroot jails. These new features have also jumpstarted the containerized ecosystem, which we will cover in Chapter 11, Enhancing the Security of Containerized Workloads.

236

Configuring Application-Specific SELinux Controls

The SELinux support for applications such as Cockpit, cron, and OpenSSH is generally provided through the SELinux policy and uses PAM integration to link SELinux controls within the application. It is, however, also possible to explicitly build in SELinux support in applications not intentionally SELinux-aware, but who support dynamic additions of logic through a modular design. As an example of this, we will look at Apache and the mod_selinux Apache module next.

Using mod_selinux with Apache Applications are often web-based, exposing their interface as either a common website or a simple web service, and executing the bulk of logic either within the web server or in backend services that the web server interacts with for the user. A web-based application has the huge advantage that end users often don't require any application or client to be installed on top of what is available by default on their device, be it a workstation, laptop, mobile, wristwatch, or smart TV. However, unlike the services discussed earlier, Apache does not run individual user sessions through PAM logins on the system. Instead, user requests are handled by the web server threads and processes themselves, which makes easy SELinux-based controls a bit harder to accomplish.

Introducing mod_selinux Apache has support for modules: dynamically loadable code that enhances the functionality of the web server, without having to rebuild the web server code itself. This modularity has given rise to the popularity of Apache, as we can see through its support for features such as PHP, introducing dynamic web applications to a server platform that was once meant to serve static content only. mod_selinux uses the same modular support, which allows the Apache web server to become SELinux-aware. Once we enable mod_selinux, we can configure Apache to switch SELinux sensitivity or even SELinux domains for running code, further isolating the behavior of the web server and allowing SELinux policies to control what the web server can do. mod_selinux also supports user mappings, allowing the Apache web server to run specific user sessions in different domains.

Before building the mod_selinux module, let's first install the necessary dependencies on the system: # yum install gcc git httpd httpd-devel redhat-rpm-config libselinux-devel

Using mod_selinux with Apache

237

Once the dependencies are installed, we can download and build the mod_selinux code. The code is available on GitHub in Kaigai's mod_selinux repository: # git clone https://github.com/kaigai/mod_selinux # cd mod_selinux # apxs -c -i mod_selinux.c

The apxs command is the Apache Extension Tool, which facilitates building and installing Apache modules. The command both compiles (-c) and installs (-i) the mod_ selinux module. We have yet to activate it in the Apache configuration though, which we accomplish by creating a new module configuration file in /etc/httpd/conf. modules.d called 99-selinux.conf (you can pick whatever name you want, but make sure it ends with the .conf suffix): LoadModule selinux_module modules/mod_selinux.so

Now, while we have now installed the module, it is not ready for consumption yet, as we have not loaded the SELinux policy for it. The mod_selinux repository contains the necessary SELinux policy code. However, it is not fully compatible with the more recent SELinux policy used by Linux distributions. We need to edit the mod_selinux.if file and remove all references to httpd_user_ script_ro_t, httpd_user_script_rw_t, and httpd_user_script_ra_t, as those types are no longer present in current SELinux policies: # sed -i '/script_r/d' mod_selinux.if

A second change – cosmetic for now – is to rename the calls from miscfiles_read_ certs to miscfiles_read_generic_certs. These are functions used in the reference policy, a different – and still the most common way – of writing SELinux policies (which we cover in Chapter 15, Using the Reference Policy), and while both functions are supported at the time of writing, the miscfiles_read_certs function is no longer recommended for use and will disappear soon: # sed -i 's/miscfiles_read_certs/miscfiles_read_generic_ certs/g' mod_selinux.if

Once we have adjusted the policy, we can build and load it. As this policy is developed using the reference policy style, the installation first requires building the module before we load it (unlike the directly loadable CIL examples we've used so far): # make -f /usr/share/selinux/devel/Makefile mod_selinux.pp # semodule -i mod_selinux.pp

238

Configuring Application-Specific SELinux Controls

With the SELinux module loaded and the mod_selinux Apache module installed, we can start configuring the Apache daemon with SELinux-specific controls.

Configuring the general Apache SELinux sensitivity The simplest configuration setting that mod_selinux supports is to configure Apache to run with a specific SELinux sensitivity. Suppose we want Apache to run with the s0-s0:c0.c100 sensitivity, then we need to adjust the Apache configuration and use the selinuxServerDomain directive. Assuming we want to adjust the sensitivity for the default welcome site, edit /etc/ httpd/conf.d/welcome.conf and add in the following code snippet:

selinuxServerDomain *:s0-s0:c0.c100

If the Apache web server uses virtual host definitions (allowing a single web server definition to manage multiple websites, based on the hostname that the client is using to access the web content), the selinuxDomainVal directive needs to be used instead of the selinuxServerDomain one. For instance, suppose the web server manages two virtual hosts, one for the apps. genfic.local domain, and the other for intranet.genfic.local, then we can assign each virtual host with its own sensitivity set like so:

DocumentRoot /srv/web/apps/htdocs ServerName apps.genfic.local selinuxDomainVal *:s0:c1,c2

DocumentRoot /srv/web/intranet/htdocs ServerName intranet.genfic.local selinuxDomainVal *:s0:c3,c4

Restart the Apache web server and validate that the setting is active: # systemctl restart httpd # ps -efZ | grep httpd_t system_u:system_r:httpd_t:s0-s0:c0.c100 ... /usr/sbin/httpd

Using mod_selinux with Apache

239

As you can see, the web server is now running with the given sensitivity. An important caveat though: the mod_selinux code does not support mcstransd, the translation daemon we covered in Chapter 3, Managing User Logins, so you cannot use human-readable sensitivity definitions such as SystemLow-SystemHigh.

Mapping end users to specific domains To map users, when logged in to a web application, to a specific domain, we need to create a user mapping file. This mapping file is then referred to using the selinuxDomainMap directive in the web server configuration. Let's first create the mapping file inside /etc/httpd/conf.d, naming it mod_ selinux.map, with the following content: test user_webapp_t:s0:c0.c100 * user_webapp_t:s0:c0,c1 __anonymous__ anon_webapp_t:s0

This mapping file contains three mappings: • The first one is for a user called test and is mapped to the user_webapp_t domain and s0:c0.c100 sensitivity. • The second one is for any successfully authenticated user and is mapped to the user_webapp_t domain and s0:c0,c1 sensitivity. • The third one is for unauthenticated users and is mapped to the anon_webapp_t domain. We can then refer to this map by adjusting the previously created snippet like so:

selinuxServerDomain *:s0-s0:c0.c100 selinuxDomainMap /etc/httpd/conf.d/mod_selinux.map

Restart the web server to apply the changes.

Changing domains based on source The mod_selinux module also supports setting the server domain value based on environment variables that we have defined elsewhere in the configuration. For instance, we can first declare the value in an environment variable when a certain condition triggers, and then tell mod_selinux that this environment variable's value is to be used for the server domain setting.

240

Configuring Application-Specific SELinux Controls

Let's make this a bit more tangible with an example. Suppose the website manages web applications for both local (internal) people, as well as for people that work from remote locations. Assuming these users enter the web server through different source IP addresses, we can use the source IP address to differentiate between the two and assign a different SELinux sensitivity value. We can do this in the Apache configuration with the SetEnvIf directive, which declares an environment variable but only if a request matches a particular condition. The condition we use is then the Remote_Addr directive, which checks the source IP address against the expression that follows. Suppose local users come from 10.10.0.0/16 and remote users from a load balancer or reverse proxy with the IP address 10.121.12.15, then we can differentiate this as follows: SetEnvIf Remote_Addr "10.10.[0-9]+.[0-9]+$" SENSITIVITY=*:s0:c0.c80 SetEnvIf Remote_Addr "10.121.12.15" SENSITIVITY=*:s0:c90 selinuxDomainEnv SENSITIVITY

It is possible to mix and match multiple mod_selinux directives. The module will use the first successful declaration, so you could use a user mapping first, and if that user mapping does not result in a hit (because the user is not declared in the map), use the environment variable, and if that fails, fall back to a default setting. All we have to do to accomplish this fallback definition is to sequentially declare the mod_ selinux directives, like so: selinuxDomainMap /etc/httpd/conf.d/mod_selinux.map selinuxDomainEnv SENSITIVITY selinuxDomainVal *:s0:c0,c1

Through these declarations, you can fine-tune the web server security using SELinux domains and sensitivities. While this should never replace the security approach within the application itself, it provides additional isolation in case an unauthorized or malicious user exploits an error within the application.

Summary

241

Summary In this chapter, we started out with an introduction to systemd and a strong focus on the service management capabilities that systemd offers. We learned how to start a service with a custom SELinux context as well as how additional files can be properly labeled upon boot. Alongside the service management, through systemd's unit files, this chapter also covered transient services and how to immediately associate the right SELinux context. Other systemd capabilities and services were touched upon as well. We saw how SELinux contexts are registered as part of the systemd journal and how to query for events using this context. We took a brief look at udev and how its rules can be used to support administrators in managing devices. One of its actions is to set the SELinux context of the device node. We then looked at D-Bus, how SELinux can be used to control the association of applications with services, and how D-Bus uses the send_msg permission to validate communications across its channels. After D-Bus, we looked at several services that use PAM to launch user contexts, and we dived into specific examples such as SSH, learning how SELinux policy developers have further fine-tuned support for these services. We finished with a look at mod_selinux, a dynamic module for Apache that enables SELinux support within Apache's configuration even though Apache itself does not have any SELinux specifics in it. In the next chapter, we will look at another SELinux-aware application, SEPostgreSQL, which extends the popular and robust PostgreSQL database with mandatory access control support through SELinux.

Questions 1. Why should you not update unit files in /usr/lib/systemd/system directly? 2. What application allows resetting the SELinux context of files during boot? 3. How can we get all log events in journald associated with a given SELinux context? 4. How can you set the SELinux label for a device node created by udev? 5. Are SELinux controls always applicable to D-Bus associations? 6. How is it possible for Apache to be SELinux-aware without Apache having any SELinux code in it?

8

SEPostgreSQL – Extending PostgreSQL with SELinux In the previous chapter, we covered a few example SELinux-aware applications: applications that know and interact with the SELinux subsystem to further enhance security within the application context. Some of these use existing policy constructs, such as Apache's mod_selinux, whereas others enhance the policy with custom classes to further fine-tune their behavior (as with D-Bus and the acquire_svc permission). With Security-Enhanced PostgreSQL (SEPostgreSQL), we get a more elaborate example of an SELinux-aware application, which uses multiple additional classes within SELinux, as well as labeling its internal database objects to further enforce security rules. In this chapter, we will learn how to apply labels within PostgreSQL, debug its enforcement rules, associate the right labels with the PostgreSQL resources, and show how this label-based security method can be used to augment specific security practices within a relational database.

244

SEPostgreSQL – Extending PostgreSQL with SELinux

In this chapter, we're going to cover the following main topics: • Introducing PostgreSQL and sepgsql • Understanding SELinux's database-specific object classes and permissions • Using MCS and MLS • Integrating SEPostgreSQL into a network

Technical requirements Check out the following video to see the Code in Action: https://bit.ly/3dDcg4Z

Introducing PostgreSQL and sepgsql PostgreSQL is a popular, featureful, and mature relational database management system. Like Apache, it also enables a modular extension of its functionalities through loadable modules. The module we will investigate is called sepgsql, shorthand for Security Enhanced PostgreSQL or SEPostgreSQL. Through sepgsql, PostgreSQL enhances itself with SELinux support for additional access controls, offering fine-grained data flow controls based on SELinux policy rules. Please be aware though that sepgsql does not implement a full mandatory access control system within PostgreSQL, as not all PostgreSQL statements will result in a policy check. While it augments the security posture of the PostgreSQL database, the module has a few limitations listed in its online documentation, available at https://www. postgresql.org/docs/10/sepgsql.html (adjust the version number in the URL as needed; the referenced document at this URL is for PostgreSQL 10, which is the version currently used within CentOS 8 and used throughout this chapter).

Reconfiguring PostgreSQL with sepgsql Before we can install sepgsql, we need to have a working PostgreSQL system at our disposal. Most Linux distributions have readily available tutorials on how to deploy PostgreSQL, which often involves creating the databases associated with it. In this chapter, we will assume that the database itself is available inside /var/lib/ pgsql/data, the default location for a CentOS-based PostgreSQL installation. The PostgreSQL configuration files are also located inside this location.

Introducing PostgreSQL and sepgsql

245

To install sepgsql, the following steps should be executed: 1. Let's first see whether the database is functioning properly by logging in as the (default) postgres superuser, and listing the currently available databases: # su postgres $ psql postgres psql (10.6) Type "help" for help. postgres=# \l List of databases Name | Owner | Encoding | -----------+----------+-------------+ postgres | postgres | UTF8 | template0 | postgres | UTF8 | template1 | postgres | UTF8 |

... ... ... ... ...

If at any point a failure occurs, check the log file inside /var/lib/pgsql/data/ log to get more information. This log file is the default log file for all PostgreSQL-related activities, as we will see when troubleshooting its SELinux support in the Troubleshooting sepgsql section. 2. Assuming PostgreSQL is working properly, let's configure it to use the sepgsql module. This module is part of the contributed modules within PostgreSQL, and is maintained by the PostgreSQL community. In CentOS, the sepgsql module is part of the postgresql-contrib package, which can be easily added to the system using yum install postgresql-contrib if it is not present yet. 3. Edit the postgresql.conf file inside /var/lib/pgsql/data and search for the shared_preload_libraries statement. By default, it will be commented out, so uncomment it and add sepgsql inside: shared_preload_libraries = 'sepgsql' # (change requires restart)

4. As mentioned, changing this parameter requires restarting the database. We will do that later, but first, we will shut down the database as our next steps will require an offline database: # systemctl stop postgresql

246

SEPostgreSQL – Extending PostgreSQL with SELinux

5. Next, we need to reconfigure all databases and enable the sepgsql-related functions. We will cover these functions in the Using sepgsql specific functions section. To enable the functions, we have to become the postgres superuser again, and for each database available, we load a specific SQL file: # su postgres $ export PGDATA=/var/lib/pgsql/data $ for DBNAME in template0 template1 postgres; do postgres --single -F -c exit_on_error=true $DBNAME < /usr/share/pgsql/contrib/sepgsql.sql > /dev/null; done

The databases listed in the example are the three default databases available in a brand-new installation. You can obtain the actual list of databases on the system through PostgreSQL's shorthand \l command, which we used earlier to check whether the database is functioning properly. 6. Let's validate whether sepgsql is working by starting the PostgreSQL database, logging in to PostgreSQL, and asking for our current context: # systemctl start postgresql # su postgres $ psql postgres ... postgres# SELECT sepgsql_getcon(); sepgsql_getcon() ----------------------------------------------------unconfined_u:unconfined_r:unconfined_t:s0-s0:c0.c1023

What we did here was to execute the freshly installed sepgsql function sepgsql_getcon(), which retrieves the current context for the session. Let's further configure the database with a test account that we can use to validate the sepgsql controls.

Creating a test account To validate whether the sepgsql controls are working, we should have a test account outside of the postgres superuser, and a local user that we can map to different SELinux contexts. As the SELinux context will heavily decide which privileges are associated with a session, we want to be able to show the impact of one context compared to the others.

Introducing PostgreSQL and sepgsql

247

First, inside PostgreSQL (with the postgres superuser), create a test account called testuser, and allow the account to authenticate with a given password: postgres=# CREATE USER testuser PASSWORD 'somepassword';

We also need to configure the database to allow password-based authentication (as the default PostgreSQL setup will use system trust or another means of authentication). To accomplish that, edit the pg_hba.conf file inside /var/lib/pgsql/data with the following settings: local all postgres peer local all testuser md5 host all testuser 127.0.0.1/32 md5 host all testuser 192.168.100.0/24 md5

The pg_hba.conf file manages the host-based authentication rules for PostgreSQL. We update it to allow password-based authentication for the testuser account (which uses md5 as an identifier) while allowing the postgres superuser to continue to authenticate using peer trust. With these changes in place, PostgreSQL allows password-based authentication of the testuser account both when the user initiates the communication over a local, socket-based interaction, as well as when a network-based communication is used. We also need to tell the SELinux policy that regular users will be allowed to connect to the PostgreSQL service: # setsebool -P selinuxuser_postgresql_connect_enabled on

While this would be sufficient for accessing the PostgreSQL service, it is not adequate to allow the regular user domain (user_t) to interact with sepgsql. To accomplish that, we need to adjust the SELinux policy so that the user_t domain is also associated with the sepgsql_client_type attribute, and that the user_r role can have the sepgsql-related types active. We do this through a small CIL policy, as follows: (typeattributeset cil_gen_require sepgsql_client_type) (typeattributeset cil_gen_require user_t) (typeattributeset cil_gen_require sepgsql_trusted_proc_t) (typeattributeset cil_gen_require sepgsql_ranged_proc_t) (typeattributeset sepgsql_client_type (user_t)) (roleattributeset cil_gen_require user_r) (roletype user_r sepgsql_trusted_proc_t) (roletype user_r sepgsql_ranged_proc_t)

248

SEPostgreSQL – Extending PostgreSQL with SELinux

It is also possible to accomplish this with a reference policy style module, as follows: policy_module(local_sepgsql, 1.0) gen_require(` role user_r; type user_t; ') postgresql_role(user_r, user_t)

Assuming we stick with the CIL-based policy, let's load the file (that is, local_ sepgsql.cil) as an SELinux policy module: # semodule -i local_sepgsql.cil

Don't forget to restart the PostgreSQL service after changing the pg_hba.conf file.

Tuning sepgsql inside PostgreSQL The sepgsql module introduces two configuration parameters that can be used to tweak sepgsql inside PostgreSQL: • The sepgsql.permissive parameter tells PostgreSQL not to enforce the SELinux policy rules inside PostgreSQL. This is similar to the permissive state of SELinux on the system, but covers the sepgsql-related functionality inside PostgreSQL alone. • The sepgsql.debug_audit parameter tells PostgreSQL to always log the SELinux-related decisions, even when they are to allow a statement to be processed. This is similar to the auditallow statements for SELinux on the system. It is very important however to understand that sepgsql is a user-space object manager, as explained in Chapter 7, Configuring Application-Specific SELinux Controls: the SELinux subsystem in the Linux kernel is not used for enforcing the access controls, only sepgsql is. The only purpose that the SELinux subsystem has is to allow PostgreSQL to query the active SELinux policy or obtain current SELinux context information. Hence, the previous configuration parameters work mostly independently of the configuration of the system. While SELinux must be active on the system, it does not need to be in enforcing mode to have sepgsql enforce the rules inside PostgreSQL, nor does a permissive SELinux system make the enforcement of sepgsql permissive as well.

Introducing PostgreSQL and sepgsql

249

The sepgsql.debug_audit parameter does have some relationship with the system policy. We can add auditallow statements to the SELinux policy to force the logging of events even when they are allowed. What the sepgsql.debug_audit parameter does is force all events to be logged, something useful for troubleshooting sepgsql, as we will see next.

Troubleshooting sepgsql Let's enable the debug statements for an individual session and reinvoke the sepgsql_getcon function again: # su postgres -c "/usr/bin/psql postgres" postgres=# SET sepgsql.debug_audit = true; SET postgres=# SELECT sepgsql_getcon(); ...

If you want to enable the configuration for the entire system, you can place the configuration inside the postgresql.conf file: sepgsql.debug_audit = true

Inside the PostgreSQL logs, we will notice the following information: STATEMENT: SET sepgsql.debug_audit = true STATEMENT: SELECT sepgsql_getcon(); LOG: SELinux: allowed { execute } \ scontext=unconfined_u:unconfined_r:unconfined_t:s0-s0:c0. c1023 \ tcontext=system_u:object_r:sepgsql_proc_exec_t:s0 \ tclass=db_procedure name="pg_catalog.sepgsql_getcon()"

The first two lines log the statements that we have executed within the session, whereas the third line is the SELinux log event related to the execution of sepgsql_getcon. The event tells us that the unconfined_t domain (source context) has attempted (and succeeded) to execute the database procedure (as indicated by the db_procedure class) labeled with the sepgsql_proc_exec_t type. The in-database function is the sepgsql_getcon function within the pg_catalog schema.

250

SEPostgreSQL – Extending PostgreSQL with SELinux

If a denial occurs, this will result in a similar event in the logs, but will also be made visible to the end user that triggered the denial, as PostgreSQL will show an error message like so: ERROR:

SELinux: security policy violation

Unlike the audit logging executed by, for instance, D-Bus (which results in USER_AVC events in the regular audit log), sepgsql will follow the log configuration of the PostgreSQL database itself, so keep a close eye out on this log file (or other log targets configured in PostgreSQL) when trying to troubleshoot sepgsql. In this simple example, you might already have noticed that the event references a database-specific class (db_procedure). In the next section, we will look into the various classes, permissions, and types associated with sepgsql and thus supported by the SELinux policy.

Understanding SELinux's database-specific object classes and permissions The sepgsql module uses several database-specific SELinux classes to fine-tune the policies and access controls. The supported classes can be listed through /sys/fs/ selinux/class or the seinfo command: # seinfo --class | grep db_ db_blob db_column db_database db_language db_procedure db_schema db_sequence db_table db_tuple db_view

These classes have an obvious relational database meaning: db_database is for database-related permissions, db_table for table permissions, db_procedure for database procedures, and so on. While not all classes are still supported by sepgsql (the db_database class has no immediate support anymore), most do have their usual mapping within the PostgreSQL database.

Understanding SELinux's database-specific object classes and permissions

251

Let's see what permissions are supported by sepgsql and how this can be used to fine-tune access controls within the database.

Understanding sepgsql permissions The access controls that sepgsql enforces are on top of the discretionary access controls already supported by PostgreSQL. Rather than using the privileges of the role or user currently acting within the database, the sepgsql module will use the context associated with the session. As we can use different SELinux contexts for sessions that are authenticated using the same database role, we can create distinct access controls within the database without associating this with the user account itself. We can, for instance, differentiate based on the initialization of the database session: a remote session might have a separate context compared to a locally launched session, or the authorizations might be unique across different Linux users even when they share the same account within the database. Important note As remote connections require the peer context to be accessible, sepgsql requires either the use of labeled IPSec, or we need to introduce fallback labeling using NetLabel and CIPSO, as seen in Chapter 5, Controlling Network Communications. We will establish such a mapping in the Integrating SEPostgreSQL in the network section, after explaining the various permission mappings.

Once logged in, a query on a table will trigger a few checks against the SELinux policy: • Any SELECT, INSERT, UPDATE, or DELETE statement on a table results in a permission check against the select, insert, update, or delete permissions within the db_table class. • When the WHERE clause lists one or more different tables, then the select privilege for those different tables is checked as well. • Furthermore, column-level permissions are checked for each referenced column, and this is checked against the permissions within the db_column class. Again, permission checks against the select permission validate the read access, whereas the update or insert permissions reflect the controls to check when the values are changed. A more elaborate overview of the supported permissions is available in the PostgreSQL sepgsql documentation.

252

SEPostgreSQL – Extending PostgreSQL with SELinux

Using the default supported types The default SELinux policy has several types readily available for use within a sepgsql setup. Most of the SEPostgreSQL configurations will not deviate from these default types, and instead rely on the category- and sensitivity-oriented controls that we touched upon in Chapter 3, Managing User Logins. To see what these default types are, what they are used for, and how to assign these labels within PostgreSQL, let's start with creating a new database called db_test: # su postgres -c "/usr/bin/psql postgres" postgres=# CREATE DATABASE db_test; CREATE DATABASE

Next, we connect to this newly created database and create a simple table, called tb_users, which has the following columns: • The user's ID, named uid • The user's name, named name • The user's email address, named mail • The user's mailing address, named address • The user's password salt and hash, named salt and phash Important note The example used is merely an example, meant to show how to approach SELinux labels and sepgsql. Proper database design and best practices for addressing password hashes and other sensitive data are well beyond the scope of this book!

As you can imagine, we will be securing some of these columns further: while the password hash should obviously be considered very sensitive, we should also make sure to properly protect the mail and address fields as this is Personally Identifiable Information (PII), which in many areas of the world is governed by specific privacy laws: postgres=# \c db_test; db_test=# CREATE TABLE tb_users(uid int primary key, name text, mail text, address text, salt text, phash text);

Understanding SELinux's database-specific object classes and permissions

253

What is now the label associated with this table? For that, we need to query the PostgreSQL internal tables/views, more specifically the pg_seclabels one: db_test=# SELECT objname,provider,label FROM pg_seclabels WHERE objname='tb_users'; objname | provider | label ----------+----------+---------------------------------------tb_users | selinux | unconfined_u:object_r:sepgsql_table_t:s0

As you can see, the table has received the sepgsql_table_t type and default sensitivity (s0). sepgsql_table_t is the default type for tables. We usually find this type used for general table support and columns. Alongside the sepgsql_table_t type, the policy has a few other table- and column-oriented types that administrators can use to differentiate the controls that sepgsql enforces:

• The sepgsql_fixed_table_t type can be used for tables or columns that can only be appended to (inserted into) but not updated. This could be for log-related tables or audit events where we want to use the sepgsql controls to further enforce this (beyond the in-database controls that could be used for this as well). • The sepgsql_ro_table_t type can be used for tables or columns that should only be read from (read-only). • The sepgsql_secret_table_t type can be used for tables or columns that cannot be accessed by regular users or sessions, and only by administrative ones. This is generally used for tables or columns that are only used through protected and/or privileged procedures. • The unpriv_sepgsql_table_t type is like the sepgsql_table_t type, but specific to tables or columns managed by admins or unconfined users that cannot be accessed by confined users. • The user_sepgsql_table_t type on the other hand is specifically constructed for tables or columns managed by confined users. This allows administrators to differentiate between user-specific tables and general tables. Let's grant the testuser account (full) access to this table and database, and add some data to the table: db_test=# GRANT ALL PRIVILEGES ON ALL TABLES IN SCHEMA public TO testuser; db_test=# INSERT INTO tb_users VALUES (1, 'Sven Vermeulen', '[email protected]', 'Some Place 10001, Somewhere', 'abc123',

254

SEPostgreSQL – Extending PostgreSQL with SELinux

'f5ba94...3'); db_test=# INSERT INTO tb_users VALUES (2, 'Lisa McCarthy', '[email protected]', 'Lisa Place 15, Someplace', 'def456', 'ba53f2...0');

If we query the data through our test user, we can see all data added to the table: db_test=> SELECT * FROM tb_users;

Let's change the type of the phash column to sepgsql_secret_table_t: db_test# SECURITY LABEL ON COLUMN tb_users.phash IS 'system_u:object_r:sepgsql_secret_table_t:s0';

This alone however will not prevent the testuser user from accessing the data. It will depend on how the testuser logs in to the database—from which context the session will be initiated. If we launch the session from an unconfined domain, then the session will still allow access to the data. Let's instead log in from a regular user session (user_t), and try to access the data again: db_test=> SELECT * FROM tb_users; ERROR: SELinux: security policy violation

Even though the user has all the privileges within the database, we notice that the policy has prevented access. We can, however, query the columns not marked as sepgsql_ secret_table_t: db_test=> SELECT uid, name, mail, address, salt from tb_users;

As the phash column is now marked as sepgsql_secret_table_t, we would still want the regular database user to be able to query if a hash matches the hash in the database, or set a new hash. This allows the database user to manage the accounts without easily leaking the password hashes. We do this through functions, which we will describe next.

Understanding SELinux's database-specific object classes and permissions

255

Creating trusted procedures PostgreSQL supports functions and procedures to facilitate isolating or combining actions within the database or on the data in a more structured and managed way. Procedures are allowed to do transactional updates in the database, but do not return a value by themselves. Functions return a value, but are not allowed to do transactional updates. In our example, we will create two functions, one to compare a hash with the stored hash (but without showing the stored hash to the database user) and another to update the stored hash. Informational note While we should be using procedures for the second function, not all PostgreSQL versions in use today support them. Support for procedures has only been included from PostgreSQL version 11 onward, whereas our examples use PostgreSQL 10.6, as that is the current version supported by CentOS 8.

Let's first create the two functions: postgres=# CREATE FUNCTION compare_hash(fuid int, fphash text) RETURNS boolean AS 'SELECT phash = regexp_replace(fphash, ''[^a-f0-9]*'', '''', ''g'') FROM tb_users WHERE uid = fuid' LANGUAGE sql; postgres=# CREATE FUNCTION set_hash(fuid int, fphash text) RETURNS int AS 'UPDATE tb_users SET phash = regexp_ replace(fphash, ''[^a-f0-9]*'', '''', ''g'') WHERE uid = fuid RETURNING uid' LANGUAGE sql;

We introduce a regular expression in the function to sanitize the input as we will be marking these functions as trusted later, and we do not want the functions to be a jumping ground for activities such as SQL injection.

256

SEPostgreSQL – Extending PostgreSQL with SELinux

Once the functions are defined, authorized users can use them to access the more protected data. Of course, we need to properly label these functions. In the default SELinux policy, the following types are available to deal with procedures and functions: • sepgsql_proc_exec_t is the type to assign to regular functions or procedures. Once executed, the procedure will run within the current context of the user, so no transition will occur. • sepgsql_trusted_proc_exec_t is the type to assign to trusted procedures or functions. Once executed, these functions will run in the sepgsql_trusted_ proc_t domain, which has access to more privileged types such as sepgsql_ secret_table_t. • sepgsql_ranged_proc_exec_t is the type to assign to a trusted procedure or function, but with an additional privilege: ranged procedures are allowed to change the current sensitivity. Ranged procedure privileges are useful to assign to a function or procedure that can access columns labeled with a category that the current context would not be able to access otherwise. Once executed, these functions and procedures will run in the sepgsql_ranged_proc_t domain. • User-managed procedures can be labeled with unpriv_sepgsql_proc_exec_t (for unconfined users) and user_sepgsql_proc_t (for confined users). These procedures and functions will continue to run in the user domain itself. To get the currently assigned label for the function, use the LIKE statement as the functions are defined (in the objname column) with variables in their name. As such, they are not always that obvious to immediately select: db_test=# SELECT objname,provider,label FROM pb_seclabels WHERE objname LIKE 'compare_hash%';

Let's mark these functions as trusted: db_test=# SECURITY LABEL ON FUNCTION compare_hash(fuid integer, fphash text) IS 'system_u:object_r:sepgsql_trusted_proc_ exec_t:s0'; db_test=# SECURITY LABEL ON FUNCTION set_hash(fuid integer, fphash text) IS 'system_u:object_r:sepgsql_trusted_proc_ exec_t:s0';

With these labels in place, the database user can execute the appropriate checks and changes even though the user has no access to the phash column itself: db_test=> SELECT compare_hash(1, 'abc123'); f

Using MCS and MLS

257

db_test=> SELECT set_hash(1, 'abc123'); 1 db_test=> SELECT compare_hash(1, 'abc123'); t

Of course, preventing unauthorized users from accessing sensitive data is not something that PostgreSQL cannot do without sepgsql. PostgreSQL can have procedures and functions marked as running with the privileges of the owner of the function or procedure, rather than the executing session. What sepgsql provides is another means to accomplish this, or offer data protection through other security models. For instance, in our example, the in-database permissions of the testuser account are still applicable, we are not granting the testuser account other privileges or escalating its privileges to a higher set – instead, we are using the SELinux labels and context information to additionally filter privileges.

Using sepgsql-specific functions The sepgsql PostgreSQL module adds a handful of functions that we can use to interact with the labeling within the database: • With sepgsql_getcon(), we can obtain the current context for the session. • With sepgsql_setcon(), we can change the context of the current session, provided that the current context has the permissions to do so, of course. • With sepgsql_restorecon(), all objects within the current database are relabeled back to the default setup. The function supports a single argument, which can be NULL, or be a reference to a file that defines the new defaults. • With sepgsql_mcstrans_in() and sepgsql_mcstrans_out(), we can interact with the mcstrans daemon (if it is running), translating from a human-readable sensitivity range to raw (_in()) or vice versa (_out()). These functions are useful when maintaining labels or defining functions that have logic included that depends on the context information.

Using MCS and MLS The most common use case for enabling the sepgsql module is to use Multi-Category Support (MCS) and Multi-Level Security (MLS) support within SELinux to fine-tune access to resources.

258

SEPostgreSQL – Extending PostgreSQL with SELinux

Limiting access to columns based on categories Suppose we use the range of category numbers from c900 to c909 to address specific PII datasets, and grant users access to these categories either by granting them direct access, or by using specific SELinux contexts to consult this data. Within the database, we could mark the PII-sensitive data with a category number within that range: db_test=# SECURITY LABEL ON COLUMN tb_users.mail IS 'system_u:object_r:sepgsql_table_t:s0:c903'; db_test=# SECURITY LABEL ON COLUMN tb_users.address IS 'system_u:object_r:sepgsql_table_t:s0:c903';

With the labels applied, a user that does not have access to this category will not be able to access the data: db_test=> SELECT sepgsql_getcon(); user_u:user_r:user_t:s0-s0:c0.c100 db_test=> SELECT uid,name,mail,address FROM tb_users; ERROR: SELinux: security policy violation;

With the category range for the user set correctly, access to the data is granted: db_test=> SELECT sepgsql_getcon(); user_u:user_r:user_t:s0-s0:c0.c100,c900.c904 db_test=> SELECT uid,name,mail,address FROM tb_users;

It is important to understand though that most domains will be allowed to switch their category set, as long as it remains within the allowed range: # semanage login -l Login Name SELinux User ... taylor user_u

MLS/MCS Range

...

s0-s0:c0.c100,c900.c903 ...

This means that, even when a user session for this user launches with a more limited category set (for instance, using the runcon command), the user will still be able to call runcon again to extend the category range, or use the sepgsql_setcon() function: db_test=> SELECT sepgsql_getcon(); user_u:user_r:user_t:s0-s0:c0.c100; db_test=> SELECT sepgsql_setcon('user_u:user_r:user_t:s0-s0:c0. c100,c900.c903');

Integrating SEPostgreSQL into the network

259

db_test=> SELECT sepgsql_getcon(); user_u:user_r:user_t:s0-s0:c0.c100,c900.c903

To remediate this, we need to have the target domain be MCS-constrained.

Constraining the user domain for sensitivity range manipulation The SELinux policy always allows reducing the category range, so a range that initially includes the c900 category can always switch to a category range that excludes this category. The rules within SELinux that grant domains the privilege to reduce their category range use dominance rules, which are basically algorithms running mathematical set expressions on the source and target set: if the target set is fully enclosed within the source set, then SELinux will allow the range transition to occur. The policy however also allows for extending the category range (if the range remains within the allowed range as defined by the SELinux configuration for the user), unless the domain itself is marked as MCS-constrained. The default MCS-constrained domains are generally those domains used for sandbox usage or virtualization, as we will see in Chapter 9, Secure Virtualization. However, we can easily add more domains. For instance, to mark the user domain as MCS-constrained, load the following CIL policy: (typeattributeset cil_gen_require mcs_constrained_type) (typeattributeset cil_gen_require user_t) (typeattributeset mcs_constrained_type (user_t))

This will prevent the user_t domain from growing its category range again.

Integrating SEPostgreSQL into the network When we use the sepgsql module in PostgreSQL, all database sessions need to have a security context associated with them. While for local communications (which use Unix domain sockets) this context is readily available, networked sessions (which are the most common) do not automatically have a context set. If the system does not participate in a labeled networking setup, as we saw in Chapter 5, Controlling Network Communications, interaction with the database will fail: $ psql -U testuser -h ppubssa3ed db_test psql: FATAL: SELinux: unable to get peer label: Protocol not available

260

SEPostgreSQL – Extending PostgreSQL with SELinux

To resolve this, the recommended approach is to start using labeled IPSec. However, we can also use NetLabel to introduce fallback labeling where needed.

Creating a fallback label for remote sessions With Linux's NetLabel and CIPSO support (as seen in Chapter 5, Controlling Network Communications) we can introduce both fallback labeling (associating a label based on the source address), as well as use full labeling for localhost communication. With full, local label support, NetLabel can pass the source context to the target if all this communication solely traverses over the loopback device (as such communication does not leave the system, allowing NetLabel to trace and support the flow from end to end and provide context information to the receiving service). Let's create the CIPSO definition for local labeling: # netlabelctl cipsov4 add local doi:2

We now create a default context for communication coming from the network (over the eth0 interface and the 192.168.100.1/24 network). It is this context that we will see when connecting to the PostgreSQL server over the network: # netlabelctl unlbl add interface:eth0 address:192.168.100.0/24 label:user_u:user_r:user_t:s0

We can now remove the default mapping rules, and add mapping rules for the different communication types: # netlabelctl map del # netlabelctl map add # netlabelctl map add # netlabelctl map add protocol:cipsov4,2

default default address:0.0.0.0/0 protocol:unlbl default address:::/0 protocol:unlbl default address:127.0.0.1

The mappings we created will allow unlabeled communication for everything (but keep in mind that we have a specific label defined for communication coming from 192.168.100.0/24) and loopback-based full labeling on the localhost.

Summary

261

Tuning the SELinux policy Next to the labeling configuration, we might also need to further fine-tune the SELinux policy for PostgreSQL. A couple of SELinux booleans are worth mentioning here: • The postgresql_selinux_transmit_client_label SELinux boolean (disabled by default) allows the postgresql_t domain to set its own session contexts. The PostgreSQL server might want to set its own session context when the server itself has database connections to other, remote databases (for instance, using PostgreSQL's Foreign Data Wrapper (FDW) support). When enabled, the client context will be passed on to the remote databases as well. • The postgresql_selinux_unconfined_dbadm SELinux boolean (enabled by default) grants administrative database privileges in sepgsql to any unconfined user domain. • The postgresql_selinux_users_ddl SELinux boolean (enabled by default) allows unprivileged users to run Data Definition Language (DDL) statements. There are database statements that create new tables, views, and so on, and will result in user-oriented types such as user_sepgsql_table_t being used. • The selinuxuser_postgresql_connect_enabled SELinux boolean (disabled by default) allows user domains to connect to the PostgreSQL daemon over the Unix domain sockets. Don't forget to persist the boolean changes (using setsebool -P) as otherwise, a system reboot will revert the settings back to their default values.

Summary The PostgreSQL database can be extended with SELinux support using the sepgsql module. The module adds label support to the various objects within a database, and checks access permissions between the session context and the target label. To obtain the session context, sepgsql relies either on purely socket-based communication, or labeled networking. In this chapter, we learned how to enable the sepgsql module and how to troubleshoot possible policy issues. We then used the various default types within an example database and used these types to show how the access controls in sepgsql work. We then used SELinux's MCS support to further handle category-based access controls. Finally, we integrated PostgreSQL in a network using fallback labeling support. In the next chapter, we will examine secure virtualization within Linux and see how SELinux contributes to the isolation of virtual guests.

262

SEPostgreSQL – Extending PostgreSQL with SELinux

Questions 1. Is SEPostgreSQL part of the default PostgreSQL technology? 2. What else needs to be additionally enabled before sepgsql can be used properly? 3. How do you set or query the labels on database objects? 4. Why are the sepgsql decision events not available in the system audit log?

9

Secure Virtualization More and more system tools have built-in support for SELinux or use SELinux's features to further harden their own service offering. When we look at virtualization, libvirt is the reigning champion as a virtualization management tool, using the QEMU and Kernel-Based Virtual Machine (KVM) hypervisors. In this chapter, administrators will learn what secure Virtualization (sVirt) is and how it is applied by the libvirt tool suite, which SELinux domains are put in place, and how sVirt uses SELinux categories to isolate guests from each other. We will study how SELinux can help reduce the risks of virtualization and understand how the SELinux policy is tuned to support virtualization services. In this chapter, we're going to cover the following main topics: • Understanding SELinux-secured virtualization • Enhancing libvirt with SELinux support • Using Vagrant with libvirt

Technical requirements Check out the following video to see the Code in Action: https://bit.ly/2T805Ug While it is possible to run the examples in this chapter on an older system, we recommend using a more modern system that has hardware support for virtualization. This will ensure higher performance during the exercises, as full emulation can severely hamper the progress, especially on older systems.

264

Secure Virtualization

To verify whether your system has hardware support for virtualization (and can therefore use the Linux KVM-based virtualization), the following command should have output: # grep -E 'svm|vmx' /proc/cpuinfo flags : fpu vme de ... vmx ...

If no output is shown, then the system does not support hardware-assisted virtualization.

Understanding SELinux-secured virtualization Virtualization is a core concept that plays a part in many infrastructural service designs. Ever since its inception in the early 1970s as a means of isolating workloads and abstracting hardware dependencies, virtualization implementations have grown tremendously. When we look at infrastructure service offerings today, we quickly realize that many cloud providers would be out of service if they could not rely on the benefits and virtues of virtualization. One of the properties that virtualization offers is isolation, which SELinux can support and augment quite nicely.

Introducing virtualization When we look at virtualization, we look at the abstraction layers it provides to hide certain resource views (such as hardware or processing power). Virtualization contributes to the development of more efficient hardware usage (which results in better cost control), centralized views on resources and systems, more flexibility in the number of operating systems that the company can deal with, standardization of resource allocation, and even improved security services. There are several virtualization types around: • Full-system emulation: Where hardware is completely emulated through software. QEMU is an open source emulation software capable of handling full-system emulation, allowing administrators and developers to run virtual platforms with different processor architectures not otherwise compatible with their own systems. • Native virtualization: Where main parts of the hardware are shared across instances, and guests can run unmodified on them. Linux's KVM, which is also supported through QEMU, is an example of this type of virtualization.

Understanding SELinux-secured virtualization

265

• Paravirtualization: Where the guest operating system uses specific APIs offered by the virtualization layer (on which unmodified operating systems cannot be hosted). Initial releases of Xen only supported paravirtualization. Using KVM with VirtIO drivers is another, more modular example. • OS-level virtualization or containerization: Where the guest uses the host operating system (kernel) but does not see the processes and other resources running on the host. Docker containers or LXC containers are examples of OS-level virtualization. • Application virtualization: Where the application runs under a specialized software runtime. A popular example here is the support for Java applications, running on the Java Virtual Machine (JVM). Many virtualization platforms support a few virtualization types. QEMU can range from full emulation to paravirtualization, depending on its configuration. When we work with virtualization layers, the following terms come up frequently: • The host is the (native) operating system or server on which the virtualization software is running. • The guest is the virtualized service (generally an operating system or container) that runs on the host. • The hypervisor is the specialized virtualization software that manages the hardware abstraction and resource-sharing capabilities of the virtualization platform. It is responsible for creating and running the virtual machines. • An image is a file or set of files that represents the filesystem, disk, or other medium assigned to a guest. • A virtual machine is the abstracted hardware or resource set in which the guest runs. We will be using these terms in this chapter, as well as in Chapter 10, Using Xen Security Modules with FLASK, and Chapter 11, Enhancing the Security of Containerized Workloads, as those chapters also cover specific virtualization implementations and how SELinux actively provides additional security controls in them. Before we embark on configuring and tuning virtualization services, let's first see what SELinux has to offer for virtualized environments.

266

Secure Virtualization

Reviewing the risks of virtualization Virtualization comes with a number of risks though. If we ask architects or other risk-conscious people about the risks of virtualization, they will talk about virtual machine sprawl, challenges related to secure or insecure APIs, the higher complexity of virtualized services, and whatnot. Going over the challenges of virtualization itself is beyond the scope of this chapter, but there are a few notable risks that play directly into SELinux's field of interest. If we can integrate SELinux with a virtualization layer, then we can mitigate these risks more proactively: • The first risk is data sensitivity within a virtual machine. Whenever multiple virtual machines are hosted together, you could have the risk that one guest is able (be it through a flaw in the virtualization software, the hypervisor's networking capabilities, or through side-channel attacks) to access sensitive data on another virtual machine. With SELinux, data sensitivity can be controlled using sensitivity ranges. Guests can run with different sensitivity ranges, guaranteeing the data sensitivity even on the virtualization layer. • Another risk is the security of offline guest images. Here, either administrators or misconfigured virtual machines might gain access to another guest image. SELinux can prevent this through properly labeled guest images and ensuring that images of offline virtual machines are typed differently from online virtual machines. • Virtual machines can also exhaust the resources on a system. On Linux systems, many resources can be controlled through the control groups (cgroups) subsystem. As this subsystem is governed through system calls and regular file APIs, SELinux can be used to further control access to this facility, ensuring that the cgroups maintained by libvirt, for instance, remain solely under the control of libvirt. • Break-out attacks, where vulnerabilities within the hypervisor are exploited to try to reach the host operating system, can be mitigated through SELinux's type enforcement as even a hypervisor does not require full administrative access to everything on the host. • SELinux can also be used to authorize access to the hypervisor, ensuring that only the right teams (through the role-based access controls) are able to control the hypervisor and its definitions.

Understanding SELinux-secured virtualization

267

• Finally, SELinux also offers improved guest isolation, which goes beyond just the guest image accesses. Thanks to SELinux's MCS implementation, guests can be separated from each other in a mandatory approach. With type enforcement, the allowed behavior of guests can be defined and controlled. This is a key capability used by hosting providers as they allow running (for them) untrusted guest virtual machines. SELinux, however, is not a full security solution for virtualization providers. One main design constraint with SELinux is that it is not dynamic if the system itself is not SELinux-aware. When we assign a type to a virtual machine, this type is generally rigid and set in stone. Virtual machines will have different behavior characteristics depending on the software running on them. A virtual machine running a web server has different behavior characteristics than one running a database or an email gateway. Although SELinux policy administrators would be capable of creating new domains for each virtual machine, this is not efficient. As a result, most SELinux policies will only offer a few domains usable by the virtual machine with broad characteristics. With libvirt, these domains are part of the sVirt solution.

Reusing existing virtualization domains When Red Hat introduced its virtualization solution, it also added SELinux support, calling the resulting technology sVirt, derived from secure virtualization. As secure virtualization as a term is hardly unique in the market, we use the term sVirt predominantly to refer to the SELinux integration within virtualization management solutions such as libvirt. With sVirt, the open source community has a reusable approach for augmenting the security posture of virtualization and containerization through SELinux. It does this through the following domains and types, which can be used regardless of the underlying virtualization platform: • The hypervisor software itself, such as libvirtd, uses the virtd_t domain. • Guests (virtual machines) that do not require any interaction with the host system and resources beyond those associated with a generic virtual machine generally use the svirt_t domain. This domain is the most isolated guest domain for full virtualization solutions. • Guests that require more interaction with the host, such as using the QEMU networking capabilities and sharing services, will use the svirt_qemu_net_t domain.

268

Secure Virtualization

• Guests that use the KVM networking capabilities and sharing services will use the svirt_kvm_net_t domain. It is very similar in permissions to svirt_qemu_ net_t but optimized for KVM. • Containerized guests, as we will see in Chapter 11, Enhancing the Security of Containerized Workloads, will use the svirt_lxc_net_t domain, whose privileges are optimized for OS-level virtualization. • Guests that require more flexible memory accesses (such as executing writable memory segments and memory stacks) will use the svirt_tcg_t domain. This flexible memory access is common for full virtualization guests whose emulation/ virtualization requires the use of a Tiny Code Generator (TCG), hence the name. • Image files that contain a guest's data will be labeled with the svirt_image_t type. • Image files that are not in use at the moment will use the default virt_image_t type. • Image files used in a read-only fashion will have the virt_content_t type assigned to them. To enable some flexibility in what the domains are allowed to do, additional SELinux booleans are put in effect, which we'll cover next.

Fine-tuning virtualization-supporting SELinux policy Use caution when toggling SELinux booleans to control the confinement of virtualization domains. Such booleans influence the SELinux policy on the host level, and cannot be used to change the access controls or privileges of individual guests. As such, when we change the value of an SELinux boolean, the change affects the permissions of all guests on that host. Let's see what the various SELinux booleans are for virtualized environments: • The staff_use_svirt boolean, if enabled, allows the staff_t user domain to interact with and manage virtual machines, as by default this is only allowed for unconfined users. • The unprivuser_use_svirt boolean, if enabled, allows unprivileged user domains (such as user_t) to interact with and manage virtual machines.

Understanding SELinux-secured virtualization

269

• With the virt_read_qemu_ga_data and virt_rw_qemu_ga_data booleans, the QEMU guest agent (which is an optional agent running inside the guests, facilitating operations such as freezing filesystems during backup routines) can read or even manage data labeled with the virt_qemu_ga_data_t type. This type, however, is not in use by default, and these SELinux booleans are disabled by default. • The virt_sandbox_share_apache_content boolean allows the guest domains to share web content. This is most commonly used for containers but is possible on guests as well if the hypervisor supports mapping host filesystems into the guest. • With virt_sandbox_use_audit enabled, this boolean allows the guest domains to send audit messages to the host's audit service. • The virt_sandbox_use_fusefs boolean grants the guest domains the privilege to mount and interact with Filesystem in Userspace (FUSE) filesystems. The virt_use_fusefs boolean allows the guests to read files on these filesystems. • If the virt_sandbox_use_netlink boolean is active, then guest domains can use Netlink system calls to manipulate the networking stack within the host. • With virt_transition_userdomain, containers can transition to a user domain (including the unconfined user domain unconfined_t). • When we enable virt_use_execmem, guests can use executable memory. • The virt_use_glusterd, virt_use_nfs, and virt_use_samba booleans allow guests to use network filesystems mounted on the host, offered through GlusterFS, NFS, and Samba respectively. Note that this does not involve mounts inside the guest itself, such as a guest that connects to an NFS server. The booleans handle interaction through filesystem mounts on the host. • Device access is also governed through some SELinux booleans, such as the virt_ use_comm boolean to interact with serial and parallel communication ports, virt_use_pcscd to allow guests to access smartcards, and virt_use_usb to grant access to USB devices. • The virt_use_rawip boolean allows guests to use and interact with raw IP sockets, allowing network interaction that circumvents some of the processing logic within the regular network stack.

270

Secure Virtualization

• With virt_use_sanlock, guests can interact with the sanlock service, a lock manager for shared storage. • When virt_use_xserver is set to true, guests can use the X server on the host. If security-sensitive operations need to be allowed for a single guest or a small set of guests, it is advisable to run those guests on an isolated host where these operations are then allowed while running the other guests on hosts where the policy does not allow these particular actions. Administrators can also use different SELinux domains for specific guests, fine-tuning the access controls for an individual virtual machine. How we can assign specific domains depends on the underlying technology of course. In the Enhancing libvirt with SELinux support section, we will introduce this for libvirt-based virtualization.

Understanding sVirt's use of MCS The SELinux domains and the mentioned types are not enough to implement proper confinement and isolation between guests. sVirt adds another layer of security by using SELinux's Multi-Category Security (MCS) extensively. Within SELinux, some domains are marked as an MCS-constrained type. When this is the case, the domain will not be able to access resources that do not have the same set of categories (or more) assigned as the current context, as it will not be able to extend their own active category set—something we saw in Chapter 8, SEPostgreSQL – Extending PostgreSQL with SELinux. The sVirt implementation ensures that the virtualization domains mentioned earlier are all marked as MCS-constrained types. This can be confirmed by asking the system which types have the mcs_constrained_type attribute set: # seinfo -amcs_constrained_type -x Type Attributes: 1 attribute mcs_constrained_type container_t netlabel_peer_t openshift_app_t openshift_t sandbox_min_t sandbox_net_t sandbox_t sandbox_web_t sandbox_x_t svirt_kvm_net_t svirt_qemu_net_t

Enhancing libvirt with SELinux support

271

svirt_t svirt_tcg_t

Through the MCS constraints, sVirt enables proper isolation between guests. Every running virtual machine (generally running as svirt_t) will be assigned two (random) SELinux categories. The images that virtual machine needs to use are assigned the same two SELinux categories. Whenever a virtual machine wants to access the wrong image, the difference in MCS categories will result in SELinux denying the access. Similarly, if one virtual machine is trying to connect to or attack another virtual machine, the MCS protections will once again prevent these actions from happening. sVirt selects two categories to allow a large number of guests to run even when there are only a few categories available. Assume that the hypervisor is running with the c10. c99 category range. That means that the hypervisor can only select 90 categories. If each guest only receives a single category, then the hypervisor can support 90 guests before allowing multiple guests to interact with each other (assuming a malicious actor found a vulnerability that allows that, of course, the hypervisor software will generally disallow such accesses as well). With two categories, however, the number of supported simultaneously running guests becomes 4,005 (the number of unique pairs in a set of 90, obtained through the formula n*(n-1)/2). Let's see what libvirt's SELinux support looks like.

Enhancing libvirt with SELinux support The libvirt project offers a virtualization abstraction layer, through which administrators can manage virtual machines without direct knowledge of or expertise in the underlying virtualization platform. As such, administrators can use the libvirt-offered tools to manage virtual machines running on QEMU, QEMU/KVM, Xen, and so on. To use the sVirt approach, libvirt can be built with SELinux support. When this is the case and the guests are governed (security-wise) through SELinux, then the sVirt domains and types are used/enforced by libvirt. The libvirt code will also perform the category selection to enforce guest isolation and will ensure that the image files are assigned the right label (image files that are in use should get a different label than inactive images files).

272

Secure Virtualization

Differentiating between shared and dedicated resources The different labels for images allow for different use cases. The image used to host the main operating system (of the guest) will generally receive the svirt_image_t label and will be recategorized with the same pair of categories as the guest runtime itself (running as svirt_t). This image is writable by the guest. When we consider an image that needs to be readable or writable by multiple guests, then libvirt can opt not to assign any categories to the file. Without categories, MCS constraints don't apply (well, they still apply, but any set of categories dominates an empty set, and as such, actions against those properly labeled files are allowed). Images that need to be mounted read-only for a guest (such as bootable media) are assigned the virt_content_t type. If they are dedicated, then categories can be assigned as well. For shared read access, no categories need to be assigned. Note that these label differences apply mainly to virtualization technologies and not container technologies.

Assessing the libvirt architecture The libvirt project has several clients that interact with the libvirtd daemon. This daemon is responsible for managing the local hypervisor software (be it QEMU/KVM, Xen, or any other virtualization software) and is even able to manage remote hypervisors. This latter functionality is often used for proprietary hypervisors that offer the necessary APIs to manage the virtual resources on the host:

Enhancing libvirt with SELinux support

273

Figure 9.1 – High-level libvirt architecture

Due to the cross-platform and cross-hypervisor nature of the libvirt project, sVirt is a good match. Instead of hypervisor-specific domains, generic (yet confined) domains are used to ensure the security of the environment.

Configuring libvirt for sVirt Most systems that support libvirt on SELinux systems will have SELinux support automatically enabled. If this is not the case, but SELinux support is possible, then all it takes is to configure libvirt to allow the SELinux security model. We map the SELinux security model in libvirt on a per-hypervisor basis.

274

Secure Virtualization

The configuration parameters related to sVirt are generally defined on a per-hypervisor basis. For instance, for the QEMU-based virtualization driver, we need to edit the / etc/libvirt/qemu.conf file. Let's look at the various parameters related to secure virtualization: • The first parameter, which defines whether sVirt is active or not, is the security_ driver parameter. While libvirt will by default enable SELinux once it detects SELinux is active, we can explicitly mark sVirt support as enabled by setting the selinux value: security_driver = "selinux"

SELinux support will by default be enabled without explicitly marking the security_driver variable in the configuration file. If you want to use libvirt without SELinux support (and consequently without sVirt), then you need to explicitly mark the security_driver setting as none: security_driver = "none"

• A second sVirt-related setting in libvirt is security_default_confined. This variable defines whether guests are by default confined (and thus associated with the sVirt protections) or not. The default value is 1, which means that the confinement is by default enabled. To disable it, you need to set it to 0: security_default_confined = 0

• Users of the libvirt software can also ask to create an unconfined guest (and libvirt allows this by default). If we set security_require_confined to 1, then no unconfined guests can be created: security_require_confined = 1

We can confirm that sVirt is running when we have a guest active on the platform, as we can then consult the label for its processes to verify that it indeed received two random categories. Let's create such a guest, using the regular QEMU hypervisor. We use an Alpine Linux ISO to boot the guest with, but that is merely an example—you can substitute it with any ISO you want: # virt-install --virt-type=qemu --name test \ --ram 128 --vcpus=1 --graphics none \ --os-variant=alpinelinux3.8 \ --cdrom=/var/lib/libvirt/boot/alpine-extended-x86_64.iso \ --disk path=/var/lib/libvirt/images/test. qcow2,size=1,format=qcow2

Enhancing libvirt with SELinux support

275

The locations mentioned are important, as they will ensure that the files are properly labeled: • In /var/lib/libvirt/boot (and /var/lib/libvirt/isos), read-only content should be placed, which will result in the files automatically being labeled with virt_content_t. • In /var/lib/libvirt/images, we create the actual guest images. When the guests are shut down, the images will be labeled with virt_image_t, but once started, the labels will be adjusted to match the categories associated with the domain. The command will create a guest called test, with 128 MB of memory and 1 vCPU. No specific graphics support will be enabled, meaning that the standard console or screen of the virtual machine will not be associated with any graphical service such as Virtual Network Computing (VNC) but will rely on a serial console definition inside the guest. Furthermore, we have the guest use a small, 1 GB disk that uses the QEMU copy-on-write (QCOW2) format. Once we have created the guest and launched it, we can check its label easily: # ps -efZ | grep test system_u:system_r:svirt_tcg_t:s0:c533,c565 /usr/bin/qemusystem-x86_64 -name guest=test,...

To list the currently defined guests, use the virsh command: # virsh list --all Id Name State -----------------------1 test running

The --all argument will ensure that even guests that are defined but are not running currently are listed as well. Important note Within libvirt, guests are actually called domains. As SELinux (and thus this book) also uses the term domain frequently when referring to the context of a process, we will be using guest as terminology when referring to libvirt's domains to keep possible confusion to a minimum.

276

Secure Virtualization

The virsh command is the main entry point for interacting with libvirt. For instance, to send a shutdown signal to a guest, you would use the shutdown argument, whereas the destroy argument will force the shutdown of the guest. Finally, to remove a definition, you would use undefine. As shown in the previous example, the guest we defined is running with the svirt_ tcg_t domain. Let's see how we can adjust the labels used by libvirt for guests.

Changing a guest's SELinux labels Once a guest has been defined, libvirt allows administrators to modify its parameters by editing an XML file representing the guest. Within this XML file, the SELinux labeling has a place as well. To view the current definition, you can use the dumpxml argument to virsh: # virsh dumpxml test

At the end of the XML, the security labels are shown. For SELinux, this could look like so:

system_u:system_r:svirt_tcg_t:s0:c533,c565 system_u:object_r:svirt_image_t:s0:c533,c565

If we want to modify these settings, we can use the edit argument to virsh: # virsh edit test

This will open the XML file in the local editor. However, once we accomplish that, we'll notice that the seclabel entries are nowhere to be found. That is because the default behavior is to use dynamic labels (hence type='dynamic') with default labels, which does not require any default definition.

Enhancing libvirt with SELinux support

277

Let's instead use a static definition, and have the guest run with the c123,c124 category pair. In the displayed XML, at the end (but still within the ... definition), place the following XML snippet:

system_u:system_r:svirt_tcg_t:s0:c123,c124

To run a guest with a different type is of course done in a similar fashion, changing svirt_tcg_t to a different type. However, keep in mind that not all types can be used regardless. For instance, the default svirt_t domain cannot be used with QEMU's full-system virtualization (as QEMU uses TCG if it cannot use KVM). Important note The default types that libvirt uses are declared inside /etc/selinux/ targeted/contexts, in the virtual_domain_context and virtual_image_context files. However, it is not recommended to change these files as they will be overwritten when SELinux policy updates are released by the distribution.

The relabel statement requests libvirt to relabel all resources for the guest according to the guest's current assigned label (relabel='yes') or not (relabel='no'). With dynamic category assignment, this will always be yes, while with static definitions both values are possible. Of course, if we want to, we can use dynamic category assignment with custom type definitions as well. For that, we declare type='dynamic' but explicitly define a label within a entity, like so:

system_u:system_r:svirt_t:s0

This will have the guest run with a dynamically associated category pair, while using a custom label rather than the default selected one.

278

Secure Virtualization

Customizing resource labels If the guest definition has relabeling active (either because it uses dynamic category assignment or on explicit request of the administrator), then the resources that the guest uses will be relabeled accordingly. Administrators can customize the labeling behavior of libvirt through the same interface we used previously: guest definition files. For instance, if we would not want libvirt to relabel the test.qcow2 file that represents the guest's disk, we could add to the XML like so:





This is useful when you want to allow the sharing of some resources across different guests, without making them readable by all guests. In such a situation, we could label the file itself with (say) svirt_image_t:s0:c123 and have the guests with category pairs always contain the category c123.

Controlling available categories When libvirt selects random categories, it does so based on its own category range. By default, MCS systems will have this range set to c0.c1023. To change the category range, we need to ensure that we launch the libvirt daemon (libvirtd) in the proper context. With systemd, we saw in Chapter 7, Configuring Application-Specific SELinux Controls, that this can be accomplished by editing the service unit file and defining the right SELinuxContext variable. Let's apply this to libvirtd as well: 1. First, copy over the system-provided libvirtd.service file to /etc/ systemd/system: # cp /usr/lib/systemd/system/libvirtd.service /etc/ systemd/system

Enhancing libvirt with SELinux support

279

2. Edit the libvirtd.service file and add the following definition: SELinuxContext=system_u:system_r:virtd_t:s0-s0:c800.c899

3. Reload the daemon definitions for systemd so that it picks up the new libvirtd. service file: # systemctl daemon-reload

4. Restart the libvirtd daemon: # systemctl stop libvirtd # systemctl start libvirtd

5. We can now start our guests again and verify that each guest is now running with a category pair within the range defined for the libvirtd daemon: # virsh start test # ps -efZ | grep virt system_u:system_r:virtd_t:s0-s0:c800.c899 /usr/sbin/ libvirtd system_u:system_r:svirt_t:s0:c846,c891 /usr/bin/qemusystem-x86_64 -name guest=test...

As we can see, the categories selected by libvirt are now within the defined range. Systems that do not use systemd can edit the SysV-style init script and use runcon: runcon -l s0-s0:c800.c899 /usr/sbin/libvirtd \ --config /etc/libvirt/libvirtd.conf --listen

Every time we launch a new guest, the libvirt code will randomly select two categories. The service will then check whether these categories are part of its own range and whether the category pair is already used or not. If any of these checks fail, libvirt will randomly select a new pair of categories until a free pair matches the requirements.

Changing the storage pool locations A very common configuration change with libvirt is to reconfigure it to use a different storage pool location. This has a slight impact on SELinux as well, as we do not have proper file context definitions for the new location.

280

Secure Virtualization

Let's see how to create a new pool location and change the SELinux configuration for it: 1. List the current storage pools to make sure the new pool name is not already taken: # virsh pool-list --all Name State Autostart ----------------------------------------------boot active yes images active yes root active yes

2. Create the target location: # mkdir /srv/images

3. Create the new storage pool with pool-define-as. In the following command, we name the pool large_images: # virsh pool-define-as large_images dir - - - - "/srv/ images" Pool large_images defined

4. Configure SELinux to label the pool properly: # semanage fcontext -a -t virt_image_t "/srv/ images(/.*)?"

5. Relabel the directory structure: # restorecon -R /srv/images

6. Have libvirt populate the directory structure: # virsh pool-build large_images

7. Start the storage pool: # virsh pool-start large_images

8. Turn on auto-start so that, when libvirtd starts, the pool is immediately usable as well: # virsh pool-autostart large_images

Using Vagrant with libvirt

281

9. We can verify that everything is functioning properly with the pool-info command: # virsh pool-info large_images

The output will show the current and available capacity for the new location. If we host the storage pool on an NFS-mounted location, then we need to enable the virt_use_nfs SELinux boolean as well. Now that we've fully grasped how to configure libvirt and SELinux for it, let's see how we can use the popular Vagrant tool with libvirt.

Using Vagrant with libvirt Vagrant is a framework to quickly spin up and manage virtual machines and is very popular within development communities. While Vagrant uses Oracle VirtualBox as the hypervisor by default, we can install a libvirt plugin to use Vagrant with libvirt, benefiting from the sVirt security offered by SELinux.

Deploying Vagrant and the libvirt plugin The Vagrant application can be installed from a single RPM file. Find the latest version at https://www.vagrantup.com/downloads.html and install it. For instance, for CentOS systems, you can use yum directly: # yum install https://releases.hashicorp.com/vagrant/2.2.9/ vagrant_2.2.9_x86_64.rpm

To install the libvirt plugin, we first need to make sure that the dependencies are installed as well. The documentation, online at https://github.com/vagrant-libvirt/ vagrant-libvirt, gives a good overview of which packages need to be installed. Do not forget this step, as dependency failures during the plugin installation are not always obvious. Once the dependencies are installed, use vagrant itself to download and install the plugin: # vagrant plugin install vagrant-libvirt

After installing the plugin, we can go forward with setting up a Vagrant box.

282

Secure Virtualization

Installing a libvirt-compatible box Vagrant uses boxes: images prepared for quick installation using Vagrant. Not all Vagrant boxes are compatible with the libvirt provider. Luckily, the Vagrant Cloud website at https://app.vagrantup.com/boxes/search?provider=libvirt allows you to quickly find compatible boxes. Suppose we want to use a Fedora image called fedora/32-cloud-base, then we can configure it as follows: 1. Create a new directory, which we will define the box configuration in, and enter this location: # mkdir vagrant # cd vagrant

2. Initialize the Vagrant box, using the fedora/32-cloud-base box definition: # vagrant init fedora/32-cloud-base

This will create an empty Vagrantfile that can be used to further configure the box. 3. Edit the Vagrantfile, and add the following code: config.vm.provider :libvirt do |libvirt| libvirt.storage_pool_name = "images" libvirt.driver = "qemu" # or kvm end

This will configure the libvirt provider to use the images directory as the default storage pool, and use the QEMU driver within libvirt. 4. Still inside the Vagrantfile, add the following code to give the box a proper name: config.vm.define :test do |test| test.vm.box = "fedora/32-cloud-base" end

The name chosen here is test, and will result in a virtual guest named vagrant_ test. 5. To launch the test guest, run the vagrant up command like so: # vagrant up --provider=libvirt

Depending on the speed of the system, this step can take a while to complete.

Summary

283

Tip Rather than calling vagrant up every time with the --provider=libvirt parameter, we can also declare the VAGRANT_ DEFAULT_PROVIDER="libvirt" environment variable and drop the command-line argument, as the environment variable will then be used.

Once the guest is up and running, you can connect to it using vagrant ssh. While you can manipulate the guest with the virsh commands, you can also use vagrant halt to shut down, or vagrant destroy followed by vagrant box remove to remove the box from the system completely.

Configuring Vagrant boxes Once a box is deployed, it is available through libvirt as a standard guest. That means that the operations we've seen before to modify its labels or tweak SELinux controls using the SELinux booleans still apply. Let's first verify that Vagrant is indeed using libvirt to launch its own boxes: # virsh list --all Id Name State ----------------------------1 vagrant_test running

Sure enough, the guest is available and titled vagrant_test. We can modify its configuration with virsh edit as well: # virsh edit vagrant_test

As long as the Vagrant box is not destroyed, the settings in libvirt will persist.

Summary Virtualization is a powerful technology whose security posture can be augmented thanks to SELinux. With sVirt, the open source community has a powerful approach to isolate guests and ensure virtual machines are only able to access the resources they should. In this chapter, we looked at virtualization and the risks associated with it. We discussed how some of these risks can be mitigated through the same set of controls that SELinux offers, such as type enforcement (limiting what guests can do) and MCS confinement (isolating guests from each other).

284

Secure Virtualization

We then covered how libvirt supports several virtualization technologies on Linux platforms and how it includes a technology called sVirt that enables SELinux integration, offering guest isolation and access controls. We saw how administrators can manipulate the sVirt logic within libvirt, such as through different domain labels or category sets. We finished the chapter with information on how to use Vagrant with libvirt. In the next chapter, we'll look at another virtualization solution, called Xen, which has adopted an SELinux-based technology for its hardening.

Questions 1. What is unique about sVirt that cannot be done with regular SELinux configuration? 2. What are the two main risks that SELinux tackles with virtualization? 3. What is the difference between virt_image_t, svirt_image_t, and virt_ content_t? 4. How do you change guest labels with libvirt? 5. How can we use Vagrant yet still benefit from sVirt?

10

Using Xen Security Modules with FLASK In Chapter 9, Secure Virtualization, we saw that libvirt is able to apply sVirt protection measures, based upon SELinux domains and category assignation, to several supported hypervisors. Xen, another popular open source hypervisor, is also supported by libvirt, but it is much more common to use Xen on its own, independent from libvirt. Xen itself has a security framework called Xen Security Modules (XSM), similar to Linux Security Modules (LSM), and an access control system called XSM-FLASK, which is their SELinux-based security framework. We'll see how Xen uses XSM, how to build Xen with XSM support, and finally, how we can apply policies to Xen domains. In this chapter, we're going to cover the following main topics: • Understanding Xen and XSM • Running XSM-enabled Xen • Applying custom XSM policies

286

Using Xen Security Modules with FLASK

Technical requirements Check out the following video to see the Code in Action: https://bit.ly/3kcCePl

Understanding Xen and XSM The Xen Project is a Linux Foundation project that maintains the Xen hypervisor. While the Xen Project manages multiple security and virtualized-related software titles, our focus is on the Xen hypervisor.

Introducing the Xen hypervisor The Xen hypervisor runs directly on top of hardware and sits in between the various virtual machines and the hardware itself. Unlike QEMU or KVM, which run as a process within Linux to offer the virtualization functionality, Xen works more independently. As a result, administrators will not see the running instances as separate processes. Instead, they need to rely on Xen commands and APIs to get more information and to interact with the Xen hypervisor. Important note As with libvirt, the Xen hypervisor uses the term domain to point to its guests. As we use the term domain frequently in SELinux to mean the SELinux type of a running process, and thus also the SELinux type of a running guest, we will use guest wherever possible. However, there will be some terminology associated with Xen where we will have to keep the domain terminology in place.

Xen always has at least one virtual guest defined, called Domain 0 (dom0). This guest manages the system and runs the Xen daemon (xend). It is through dom0 that administrators will create and operate virtual guests running within Xen. These regular guests are unprivileged, and therefore abbreviated as domU—unprivileged domains. When administrators boot a Xen host, they boot into Xen's dom0 instance, through which they then further interact with Xen. The Linux kernel has included support for running both within dom0 as well as domU for quite some time now (with complete support, including backend drivers, since Linux kernel 3.0). Let's use an existing Linux deployment to install Xen, and use this existing deployment as Xen's dom0 guest.

Understanding Xen and XSM

287

Installing Xen While many Linux distributions offer Xen out of the box, it is very likely that these deployments do not support XSM (which we will enable in the Running XSM-enabled Xen section). So, rather than fiddling with prebuilt Xen environments first, we want to build it from source as released by the Xen Project immediately. Before we start using Xen, let alone its XSM support, we first need to make sure that we are running with a Xen-enabled Linux kernel.

Running with a Xen-enabled Linux kernel The Linux kernel on the system must have support for running (at least) inside a dom0 guest. Without this support, not only will the dom0 guest not be able to interact with the Xen hypervisor, it will also not be able to boot the Xen hypervisor itself (the Xen-enabled kernel needs to bootstrap the Xen hypervisor before launching itself as the dom0 guest). If you build your own Linux kernel, you need to configure the kernel with the settings as documented at https://wiki.xenproject.org/wiki/Mainline_Linux_ Kernel_Configs. Some Linux distributions provide more in-depth build instructions (such as Gentoo at https://wiki.gentoo.org/wiki/Xen). On CentOS, however, out-of-the-box Xen support is currently missing from the last release (as CentOS focuses more on libvirt and related technologies for its virtualization support). Luckily, the community offers well-maintained Linux kernel builds that do include Xen support, through the kernel-ml package. Let's install this kernel package: 1. Enable the Enterprise Linux Repository (ELRepo), which introduces several other, community-driven repositories: # yum install elrepo-release

2. Install the kernel-ml package, which will install the most recent Linux kernel, with a configuration that includes Xen support. We simultaneously enable the elrepo-kernel repository, through which this package is made available: # yum install --enablerepo=elrepo-kernel kernel-ml

3. Generally, the Linux boot loader will be reconfigured to include these new kernels. If not, or you want to make sure that the kernel is properly detected, the following command can be used to regenerate the Grand Unified Bootloader (GRUB2) configuration file: # grub2-mkconfig -o /boot/grub2/grub.cfg

288

Using Xen Security Modules with FLASK

Of course, if your system uses a different boot loader, different instructions apply. Consult your Linux distribution's documentation for more information on how to configure the boot loader. 4. Reboot the system using the newly installed kernel: # reboot

If all goes well, you will now be running with a Xen-compatible kernel. That, of course, does not mean that Xen is active, but merely that the kernel can support Xen if it is needed. Let's now move forward with building the Xen hypervisor and related tooling.

Building Xen from source The Xen hypervisor and tools have dependencies on various programs and libraries, and not all tools and libraries are properly detected as dependencies while building Xen from source. Let's first install these dependencies: 1. Enable the PowerTools repository: # dnf config-manger --set-enabled PowerTools

2. Install the dependencies supported by the CentOS repositories: # yum install gcc xz-devel python36-devel acpica-tools uuid-devel ncurses-devel glib2-devel pixman-devel yajl yajl-devel zlib-devel transfig pandoc perl-Pod-Html git glibc-devel.i686 patch libuuid-devel

3. Install the dev86 package. At the time of writing, this package is not yet available for CentOS 8 so we deploy the version from CentOS 7 instead: # yum install https://download-ib01.fedoraproject.org/ pub/epel/7/x86_64/Packages/d/dev86-0.16.21-2.el7.x86_64. rpm

With the dependencies now installed, let's download the latest Xen and build it: 1. Go to https://xenproject.org/downloads/ and go to the last Xen Project release. 2. At the bottom of the page, download the latest archive.

Understanding Xen and XSM

289

3. Unpack the downloaded archive on the system: $ tar xvf xen-4.13.1.tar.gz

4. Enter the directory the archive is unpacked in: $ cd xen-4.13.1

5. Configure the sources for the local system. At this point, no specific arguments need to be passed on: $ ./configure

6. Build the Xen hypervisor and associated tools: $ make world

7. Install the Xen hypervisor and tools on the system: # make install

8. Reconfigure the boot loader. This should automatically detect the Xen binaries and add the necessary boot loader entries: # grub2-mkconfig -o /boot/grub2/grub.cfg

9. Configure the system to support libraries installed in /usr/local/lib: # echo "/usr/local/lib" > /etc/ld.so.conf.d/local-xen. conf # ldconfig

10. Create equivalence rules for the subdirectories in /usr/local so that SELinux file contexts are correctly applied: # semanage fcontext -a -e /usr/local/bin /usr/bin # semanage fcontext -a -e /usr/local/sbin /usr/sbin

11. Relabel the files inside /usr/local: # restorecon -RvF /usr/local

290

Using Xen Security Modules with FLASK

12. The result of these steps is that Xen is ready to be booted on the system. The boot loader will not use the Xen-enabled kernel by default though, so during reboot, it is important to select the right entry. Its title will contain with Xen hypervisor: # reboot

13. After rebooting into the Xen-enabled system, all we need to do is to start the Xen daemons: # # # #

systemctl systemctl systemctl systemctl

start start start start

xencommons xendomains xendriverdomain xen-watchdog

14. To verify that everything is working as expected, list the currently running guests: # xl list Name Domain-0

ID 0

Mem 7836

VCPUs 4

State r-----

Time(s) 46.2

The listing should contain a single guest, named Domain-0, which is the guest you just executed the xl list command in. 15. Finalize the installation by ensuring that the previously started daemons are started at boot: # # # #

systemctl systemctl systemctl systemctl

enable enable enable enable

xencommons xendomains xendriverdomain xen-watchdog

Before we move on to XSM, let's also create a guest inside Xen (as a domU) so that we can associate policies with it later, in the Using XSM labels section.

Creating an unprivileged guest When the Xen hypervisor is active, the operating system through which we interact with Xen is called dom0 and is the (only) privileged guest that Xen supports. The other guests are unprivileged, and it is the interaction between these guests and the actions taken by these guests that we want to isolate and protect further with XSM.

Understanding Xen and XSM

291

Let's first create a simple, unprivileged guest to run alongside the privileged dom0 one. We use Alpine Linux in this example, but you can easily substitute this with other distributions or operating systems. This example will use the ParaVirtualized (PV) guest approach, but Xen also supports Hardware Virtual Machine (HVM) guests: 1. Download the ISO for the Alpine Linux distribution, as this distribution is more optimized for low memory consumption and lower (virtual) disk size requirements. Of course, you are free to pick other distributions as well if your system can handle it. We pick the release optimized for virtual systems from https://www. alpinelinux.org/downloads/ and store the ISO on the system in /srv/ data. 2. Mount the ISO on the system so that we can use its bootable kernel when creating an unprivileged guest in our next steps: # mount -o loop -t iso9660 /srv/data/alpine-virt3.8.0-x86_64.iso /media/cdrom

3. Create an image file, which will be used as the boot disk for the virtual guest: # dd if=/dev/zero of=/srv/data/a1.img bs=1M count=3000

4. Next, create a configuration file for the virtual guest. We call the file a1.cfg and place it in /etc/xen: # Alpine Linux PV DomU # Kernel paths for install kernel = "/media/cdrom/boot/vmlinuz-virt" ramdisk = "/media/cdrom/boot/initramfs-virt" extra = "modules=loop,squashfs console=hvc0" # Path to HDD and ISO file disk = [ 'format=raw, vdev=xvda, access=w, target=/srv/data/ a1.img', 'format=raw, vdev=xvdc, access=r, devtype=cdrom, target=/srv/data/alpine-virt-3.8.0-x86_64.iso' ] # DomU settings memory = 512 name = "alpine-a1"

292

Using Xen Security Modules with FLASK vcpus = 1 maxvcpus = 1

5. Boot the virtual guest using the xl create command: # xl create -f /etc/xen/a1.cfg -c

The -c option will immediately show the console to interact with, allowing you to initiate and complete the installation of the operating system in the guest. 6. When the guest needs to reboot, use shutdown instead, and edit the configuration file. Remove the line referring to the ISO to prevent the guest from booting into the installation environment again. 7. To launch the guest again, use the xl create command again. If the guest installation finishes and you no longer need to have access to the console, drop the -c option: # xl create -f /etc/xen/xa1.cfg

8. We can confirm that the virtual guest is running with xl list: # xl list Name Domain-0 alpina-a1

ID 0 1

Mem VCPUs 7836 4 128 1

State r-----b----

Time(s) 99.4 2.5

With Xen, guests are launched with the create subcommand and shut down with the shutdown (graceful) or destroy subcommands. With these steps behind us, we now have a working Xen installation and a running guest. It's time to learn what Xen has to offer us from a security perspective.

Understanding Xen Security Modules In Chapter 1, Fundamental SELinux Concepts, we learned that SELinux is implemented through a Linux subsystem called Linux Security Modules (LSM). Xen has borrowed this idea and has a similar approach to its own security measures.

Running XSM-enabled Xen

293

With Xen Security Modules (XSM), Xen makes it possible to define and control actions between Xen guests, and between a Xen guest and the Xen hypervisor. Unlike the Linux kernel though, where several mandatory access control frameworks exist that can plug into the LSM subsystem, Xen currently only has a single module available for XSM, called XSM-FLASK. FLASK stands for Flux Advanced Security Kernel and is the security architecture and approach that SELinux also uses for its own access control expressions. With XSM-FLASK, developers and administrators can do the following: • Define permissions and fine-grained access controls between guests • Define limited privilege escalation for otherwise unprivileged guests • Control direct hardware and device access from guests on a policy level • Restrict and audit activities executed by privileged guests While XSM-FLASK uses SELinux-like naming conventions (and even SELinux build tools to build the policy), the XSM-FLASK-related settings are independent of SELinux. If dom0 is running with SELinux enabled (and there is no reason why it shouldn't), its policy has nothing to do with the XSM-FLASK policy. The labels that XSM-FLASK uses will also not be visible for regular Linux commands running inside the guests (and thus also dom0). As the running guests are not shown as processes within the system, they do not have an SELinux label at all, only an XSM-FLASK label (if enabled). Hence, Xen cannot benefit from the sVirt approach, as documented in Chapter 9, Secure Virtualization.

Running XSM-enabled Xen Switching from a regular Xen deployment to an XSM-enabled Xen deployment is a matter of rebuilding Xen with XSM support and rebooting the system. Xen comes with an out-of-the-box policy that can be readily applied, which we will use as part of our XSM endeavor.

Rebuilding Xen with XSM support Let's rebuild the Xen hypervisor and tools on the system with XSM support: 1. Clean up the previous build by running the make clean command inside the build directory (xen-4.13.1 in our example): $ make clean

294

Using Xen Security Modules with FLASK

2. Inside the build directory, go to the xen directory: $ cd xen

3. Launch the Xen configuration using make menuconfig: $ make menuconfig

4. Navigate to the XSM setting and enable the XSM-related parameters: Common Features ---> [*] Xen Security Modules support [*] FLux Advanced Security Kernel support [*] Compile Xen with a built-in FLAS security policy [*] SILO support Default XSM implementation (FLux Advanced Security Kernel)

5. Go back to the main build directory (xen-4.13.1 in our example): $ cd ..

6. Rebuild the Xen hypervisor and tools: $ ./configure $ make world

7. Install the updated Xen build on the system: # make install

This will not only update the tools but will also provide an updated Xen kernel and an XSM policy inside /boot (named xenpolicy-4.13.1). 8. Reconfigure the boot loader with the new Xen build, ensuring that the XSM policy is also loaded with it: # grub2-mkconfig -o /boot/grub2/grub.cfg

9. Reboot the system: # reboot

Running XSM-enabled Xen

295

10. Once rebooted, we can verify that the XSM policy is loaded and used by querying Xen for the labels associated with the running guests: # xl list -Z Name Domain-0 alpina-a1

ID 0 1

... ... ...

Security Label system_u:system_r:dom0_t system_u:system_r:domU_t

If the xl list command, given the -Z argument, lists the security labels, then Xen is running with an XSM policy active. Let's see where these labels are used.

Using XSM labels When Xen boots with XSM support and has its default policy active, the following types can be used by guests: • The dom0_t type is reserved for the privileged guest. • The domU_t type is the default type to use for unprivileged guests. • The isolated_domU_t type is the type to assign to unprivileged guests that should not be able to interact with other unprivileged guests, only with the privileged dom0 one. • The prot_domU_t type is meant for guests that will be prevented from starting if the XSM policy boolean prot_doms_locked is set. • The nomigrate_t type is applied to guests that are not allowed to be migrated from one Xen host to another. Internally, this prevents the dom0 guest from accessing the guest's memory once booted. There are a few other types also available inside the XSM policy that are not meant for regular guests themselves: • The dm_dom_t type is assigned to the device model guest. This is a special, privileged guest that represents the hardware virtualized for an HVM-type guest, without jeopardizing dom0. • The xenstore_t type is assigned to the xenstore stub guest. This is a special, privileged guest that provides support for unprivileged guests to access their virtualized resources, without jeopardizing dom0. • The nic_dev_t type is assigned to hardware devices that can be used in passthrough mode (meaning domU guests can directly interact with these hardware devices rather than going through the privileged guests).

296

Using Xen Security Modules with FLASK

These stub guests (stub domains or stubdoms as they are called in Xen) are a way for Xen to further increase its security posture, as privileged operations that cannot be prevented are more isolated from dom0. If at any point a security vulnerability can be exploited in these privileged services, they do not necessarily affect dom0 and, with a proper XSM policy, can even be mitigated fully. Assigning one of these labels to a guest is a matter of editing the guest's configuration file inside /etc/xen and adding in the seclabel configuration parameter: seclabel = 'system_u:system_r:isolated_domU_t'

Once configured and rebooted (using xl create), the new label will be visible when querying the running guests: # xl list -Z Name Domain-0 alpina-a1

ID 0 1

... Security Label ... system_u:system_r:dom0_t ... system_u:system_r:isolated_domU_t

Applying the right label to the guest is the most common use case (as it effectively handles the access control and protection measures we seek from the XSM implementation), but other operations are supported as well.

Manipulating XSM As with SELinux, several activities can be executed to further manipulate the XSM subsystem or the active policy.

Defining the state, ranging from disabled to enforcing When Xen boots, we can add a kernel parameter called flask, which can be set to one of the following values: • With flask=enforcing, we ensure that XSM is active, enforcing the policy between its guests and resources, and that the enforcement is immediate (no delayed activation). • With flask=permissive, XSM will load the policy, but XSM will not enforce the rules set in the policy. This is obviously meant for development purposes and behaves similarly to SELinux's permissive mode.

Running XSM-enabled Xen

297

• With flask=late, XSM will not enforce any access controls until a policy is loaded, after which the policy is enforced. This allows administrators to boot with XSM active, but only to load and apply a policy when the administrator deems it ready. • With flask=disabled, XSM will not enforce any access controls nor load the policy. This parameter can be set either directly when booting (from the boot loader) or through the boot loader configuration on the system. For instance, with GRUB2, we can edit / etc/default/grub and add or modify the following parameter: GRUB_CMDLINE_XEN_DEFAULT="flask=enforcing"

Don't forget to regenerate the GRUB2 configuration file: # grub2-mkconfig -o /boot/grub2/grub.cfg

As with SELinux, we can also manipulate the state of XSM through the command line. With xl getenforce, we can query the current state: # xl getenforce Enforcing

The xl setenforce command can be used to switch to another state: # xl setenforce permissive

These commands have nothing to do with the SELinux configuration within dom0: switching Xen from permissive mode to enforcing or vice versa is specific to Xen and has no impact on the SELinux settings inside dom0.

Querying XSM logs Like SELinux, XSM also uses AVC logging to provide feedback to the administrator about the decisions it has taken. With xl dmesg, we can query this log information (alongside the other Xen output logging): # xl dmesg ... (XEN) avc: granted { setenforce } for domid=0 scontext=system_u:system_r:dom0_t tcontext=system_u:system_r:security_t tclass=security

298

Using Xen Security Modules with FLASK

Not all granted operations will be logged, but denied operations will always result in an AVC entry. The AVC entries themselves are fully formatted like SELinux AVC entries, allowing administrators to use SELinux tools such as audit2allow to generate XSM policies.

Using XSM booleans The default policy enabled by Xen has two booleans that can be toggled: • The guest_writeconsole boolean, which defaults to 1 (on), allows guests to access and write to the Xen console. • The prot_doms_locked boolean, which defaults to 0 (off), will disallow prot_ domU_t guests from launching if enabled. While no subcommand is available for the xl command to query and set XSM booleans, two other commands are installed on the system to accomplish this – flask-get-bool and flask-set-bool: • With flask-get-bool, we can query the current state of a boolean, or list all booleans with their current value: # flask-get-bool -a guest_writeconsole: 1 prot_doms_locked: 0

• The flask-set-bool command is used to toggle booleans: # flask-set-bool prot_doms_locked 1

This is very similar to SELinux's getsebool and setsebool commands.

Querying the XSM policy The XSM policy file (xenpolicy-4.13.1) is quite similar to an SELinux policy file. As a result, we can use the SELinux tools to query this file and learn more about the policy: • With seinfo, we can query statistics about the policy, view which classes are supported, the constraints that are enabled within, and more. The only query that fails is listing the types supported within the policy: $ seinfo --all ./xenpolicy-4.13.1

Applying custom XSM policies

299

• With sesearch, we can query the XSM policy rules themselves, for instance, to list all allow rules: $ sesearch -A ./xenpolicy-4.13.1

When we discuss analyzing SELinux policies in Chapter 13, Analyzing Policy Behavior, we will get familiar with other tools that can also be used to analyze XSM policy files.

Labeling hardware resources With the flask-label-pci command, administrators can label specified PCI devices with a given type. This approach allows administrators to mark certain devices for passthrough access by unprivileged guests. For instance, to label the PCI device with address 3:2:0 with the nic_dev_t type, use the following # flask-label-pci 0000:03:02.0 system_u:object_r:nic_dev_t

As you might guess from the name, this type is initially defined for passthrough access to network devices but can be used for other PCI hardware as well.

Applying custom XSM policies Xen also allows administrators to build and use their own, custom policy. The default policy for Xen is available inside the tools/flask/policy directory within the Xen build directory. For instance, the policy rules for the dom0 guest are available inside modules/dom0.te. Important note Adjusting the Xen XSM policy is beyond the scope of this chapter. You will find instructions on how to create SELinux policies using the reference policy-style method in Chapter 15, Using the Reference Policy. The Xen XSM policy is based upon this style.

Building a custom policy is a matter of updating these files (make a backup before you do) and then rebuilding the policy itself: $ make

300

Using Xen Security Modules with FLASK

The result of the policy build is a new xenpolicy-4.13.1 file. This file can be loaded directly using the xl loadpolicy command: # xl loadpolicy /path/to/xenpolicy-4.13.1

This command is similar to the flask-loadpolicy command: # flask-loadpolicy /path/to/xenpolicy-4.13.1

If, after testing, the policy is deemed ready to be used continuously, copy it over to /boot so that it is automatically picked up at the next boot as well.

Summary The Xen hypervisor is quite different from the QEMU and KVM hypervisors, which are more readily used in libvirt. SELinux support for Xen is also different than sVirt as the SELinux subsystem can only be active inside Xen guests, and SELinux does not see other guests. Xen has resolved that by implementing its own SELinux copy as XSM-FLASK and has integrated the appropriate support for the XSM-FLASK labels in its own tooling. In this chapter, we've learned how to apply our own types to Xen guests, toggle the XSM state, toggle XSM booleans, and even how we can build and load our own XSM-FLASK policy. In the next chapter, we'll look at container workloads and how SELinux can help administrators to further harden and secure their container runtimes. We will see how sVirt can be applied to container runtimes, and how the tooling deals with SELinux support.

Questions 1. Why doesn't the regular SELinux subsystem govern Xen guests? 2. How are labels assigned to Xen guests? 3. What are the common Xen commands that deal with XSM labels? 4. How can administrators load a custom policy for testing purposes?

11

Enhancing the Security of Containerized Workloads Container platforms and management frameworks provide application-level abstraction to administrators and developers. Lightweight container frameworks allow for rapid development and deployment of new applications, whereas heavier container platforms allow for optimal resource consumption and highly resilient hosting platforms. SELinux plays a vital role in many of these frameworks and platforms, ensuring that untrusted containers cannot escape or interact with resources they are not supported to interact with. In this chapter, we look at how SELinux is supported, ranging from systemd-nspawn to podman (and Docker), and finally in larger environments with Kubernetes. We also learn how to create custom SELinux domains for containers using the udica utility.

302

Enhancing the Security of Containerized Workloads

In this chapter, we're going to cover the following main topics: • Using SELinux with systemd's container support • Configuring podman • Leveraging Kubernetes' SELinux support

Technical requirements Check out the following video to see the Code in Action: https://bit.ly/34aHOfl

Using SELinux with systemd's container support In Chapter 7, Configuring Application-Specific SELinux Controls, we introduced systemd as an SELinux-aware application suite, capable of launching different services with configurable SELinux contexts. Besides service support, systemd has quite a few other features up its sleeve. One of these features is systemd-nspawn. With systemd-nspawn, systemd provides container capabilities, allowing administrators to interact with systemd-managed containers in an integrated way, almost as if these containers were services themselves. It uses the same primitives as LXC from the Linux Containers project (which was the predecessor of the modern container frameworks) and Docker, based upon namespaces (hence the n in nspawn). Informational note The Linux Containers project has a product called LXC that combines several isolation and resource management services within the Linux kernel, such as control groups (cgroups) and namespace isolation. cgroups allow for capping or throttling resource consumption in the CPU, memory, and I/O, whereas namespaces allow for hiding information and limiting the view on system resources. Early versions of Docker were built upon LXC, although Docker has since embraced the Linux services itself directly without using LXC.

Using SELinux with systemd's container support

303

SELinux-wise, the software running inside the container might not have a correct view on the SELinux state (depending on the container configuration) as the container is isolated from the host itself. SELinux does not yet have namespace support to allow containers or other isolated processes to have their own SELinux view, so if a container has a view on the SELinux state, it should never be allowed to modify it. Now, unlike Docker, podman, and Kubernetes, which can use the sVirt approach we saw in Chapter 9, Secure Virtualization, the systemd-nspawn approach does not support this technology. Informational note The systemd-nspawn command might not be installed by default. On CentOS, Debian, and related distributions, the package that provides this tool is called systemd-container. Other distributions such as Gentoo and Arch Linux have it installed as part of the default systemd installation.

Let's see how systemd-nspawn works and what its SELinux support looks like.

Initializing a systemd container To create a systemd container, we need to create a place on the filesystem where its files will be stored, and then call systemd-nspawn with the correct arguments. To prepare the filesystem, we can download prebuilt container images, or create one ourselves. Let's use the Jailkit software, as used in Chapter 7, Configuring Application-Specific SELinux Controls, and build a container from it: 1. First, create the directory the container runtimes will be hosted in: # mkdir /srv/ctr

2. Edit the /etc/jailkit/jk_init.ini file and include the following section: [nginx] comment = nginx runtime paths = /usr/sbin/nginx, /etc/nginx, /var/log/nginx, / var/lib/nginx, /usr/share/nginx, /usr/lib64/nginx, /usr/ lib64/perl5/vendor_perl users = root,nginx groups = root,nginx includesections = netbasics, uidbasics, perl

This section tells Jailkit what it should copy into the directory, and which users to support.

304

Enhancing the Security of Containerized Workloads

3. Execute the jk_init command to populate the directory: # jk_init -v -j /srv/ctr/nginx nginx

4. Finally, start the container using systemd-nspawn: # systemd-nspawn -D /srv/ctr/nginx /usr/sbin/nginx \ -g "daemon off;"

As Nginx will by default attempt to run as a daemon, the container would immediately stop as it no longer has an active process. By launching with the daemon off option, nginx will remain in the foreground, and the container can continue to work.

Using a specific SELinux context When we launch a container directly, this container will run with the SELinux context of the user. We can, however, pass on the target context for the container using command-line arguments: • The --selinux-context= option (-Z for short) allows the administrator to define the SELinux context for the runtime processes of the container. • The --selinux-apifs-context= option (-L for short) allows the administrator to define the SELinux context for the files and filesystem of the container. The SELinux types that can be used here, however, need to be carefully selected. The processes running inside a container cannot perform any type of transitions, so regular SELinux domains are often not feasible to use. Taking our Nginx example again, the httpd_t domain cannot be used for this container. We can use the SELinux types that the distribution provides for container workloads. Recent CentOS versions will use a domain such as container_t (which was previously known as svirt_lxc_net_t) and a file-oriented SELinux type, container_file_t. While this domain does not hold all possible privileges needed for any container, it provides a good baseline for containers. Let's use this type for our container: 1. First, we need to extend the container_t privileges with some additional rights for the nginx daemon. Create a CIL policy file with the following content: (typeattributeset cil_gen_require container_t) (typeattributeset cil_gen_require container_file_t) (typeattributeset cil_gen_require http_port_t)

Using SELinux with systemd's container support

305

(typeattributeset cil_gen_require node_t) (allow container_t container_file_t (chr_file (read open getattr ioctl write))) (allow container_t self (tcp_socket (create setopt bind listen accept read write))) (allow container_t http_port_t (tcp_socket (name_bind))) (allow container_t node_t (tcp_socket (node_bind))) (allow container_t self (capability (net_bind_service setgid setuid)))

2. Load this file as a new SELinux module: # semodule -i custom_container.cil

3. Relabel the files of the container with the container_file_t SELinux type: # chcon -R -t container_file_t /srv/ctr/nginx

4. Launch the container with the appropriate labels: # systemd-nspawn -D /srv/ctr/nginx \ -Z system_u:system_r:container_t:s0 \ -L system_u:object_r:container_file_t:s0 \ /usr/sbin/nginx -g "daemon off;"

Whenever a container is launched, it remains attached to the current session. We can of course create service files that launch the containers in the background, or use session management services such as screen or tmux. A more user-friendly approach, however, is to use machinectl.

Facilitating container management with machinectl The machinectl command allows administrators to manage containers or even virtual machines more easily through systemd. For containers, machinectl will use systemd-nspawn. Let's use this machinectl command to download, start, and stop a container: 1. First, download a ready-to-go container image with the pull-tar argument and prepare it on the system: # machinectl pull-tar https://nspawn.org/storage/ archlinux/archlinux/tar/image.tar.xz archlinux

306

Enhancing the Security of Containerized Workloads

We can also download the archive manually, and then import it using machinectl import-tar: # machinectl import-tar archlinux.tar.xz

2. List the available images with the list-images argument: # machinectl list-images

3. We can now clone this image and launch the container: # machinectl clone archlinux test # machinectl start test

4. To access the container environment, use the shell argument: # machinectl shell test

5. We can shut down the container using the poweroff argument: # machinectl poweroff test

When we use machinectl, the containers will run in the unconfined_service_t SELinux domain. There is currently no way to override this. Luckily, we have other tools available to facilitate container management that do have more significant built-in SELinux support, such as Docker and podman.

Configuring podman The podman utility is the default container management utility on CentOS 8 and other distributions derived from Red Hat Enterprise Linux. Other distributions such as Gentoo can also easily get access to podman by installing libpod.

Selecting podman over Docker When we compare podman with Docker, we might not see a big difference when we are simply using it for basic container management operations. The commands are very similar, and podman even has a Docker compatibility layer that facilitates the usage of podman for administrators who are used to working with Docker.

Configuring podman

307

Under the hood though, there are quite a few differences. For one, podman is a daemon-less container management system, which allows end users to easily run containers within their confined space. The libpod project also uses different design principles and supports a different container runtime, which supports the Open Container Initiative (OCI)-based definitions, called the Container Runtime Interface for OCI (CRI-O). Let's use podman to deploy a PostgreSQL container on the system: 1. First, we need to find the appropriate container. We can use the podman search command for this: # podman search postgresql

2. Of the various PostgreSQL containers listed, we pick the Bitnami one: # podman pull docker.io/bitnami/postgresql

This command will download the container base image to the system, storing the files in /var/lib/containers/storage. 3. We can now launch a container, assign a password to the PostgreSQL superuser (postgres), and make sure that the PostgreSQL port (5432) is made available to the system: # podman run -dit --name postgresql-test \ -e POSTGRESQL_PASSWORD="pgsqlpass" \ -p 5432:5432 postgresql

This command will create a container definition, based on the container base we've just downloaded, and start it on the system. 4. We can use psql to validate that the database runs: # psql -U postgres -h localhost

5. When we're done with the container, we can stop it (using podman stop), which keeps the current container information, allowing us to revive it again later (using podman start) or remove the container from the system completely: # podman rm postgresql-test

308

Enhancing the Security of Containerized Workloads

6. Removing the container removes the container runtime, but the base container image remains on the system: # podman images

This allows us to quickly start another container without having to download the files again. Using containers is a fast and effective way to quickly install and deploy software on the system. Additionally, SELinux provides some additional protections to make sure that these containers do not misbehave.

Using containers with SELinux When we look at the active runtime, we will notice that SELinux is already confining these containers in a way we understand: # ps -efZ | grep postgres system_u:system_r:container_t:s0:c182,c609 ... /opt/bitnami/ postgres -D ...

The running processes have two categories assigned and are executing in the container_t SELinux domain. This is the sVirt approach we saw in Chapter 9, Secure Virtualization. Unlike virtual machines though, containers are often used in a more transient way: when a new version of the container base is released, the containers are scrapped, and new ones are started. Virtual machines often undergo in-system upgrades, and thus have a longer lifespan. The transient approach with containers also means that we need to provide data persistence in a different way. The approach that most containers use is to allow mapping locations from the host into the container environment. Let's use podman to map a location outside the container to the /bitnami/ postgresql location inside the container, as needed by the PostgreSQL container: 1. First, create the location where we want to store the PostgreSQL data(base) on the host: # mkdir -p /srv/db/postgresql-test

2. Next, change the ownership of this location to the user with user ID 1001 (the user ID that the container uses internally): # chown -R 1001:1001 /srv/db/postgresql-test

Configuring podman

309

3. Now start the container, creating a mapping from this location to the container: # podman run -dit --name postgresql-test \ -e POSTGRESQL_PASSWORD="pgsqlpass" \ -v /srv/db/postgresql-test:/bitnami/postgresql:Z \ -p 5432:5432 postgresql

This will have the PostgreSQL data stored in /srv/db/postgresql-test. If we later delete the container and create a new one (for instance, because an update for the container base has been made available), the database itself is not affected. The mapping itself contains an SELinux-specific variable, namely the trailing :Z. If we were to omit this from the mapping, then the location would still be made accessible inside the container. However, the PostgreSQL runtime would not be able to use it. Important note The use of :Z in directory mappings (or volume mounts as they are also often called) is the most frequently forgotten option that system administrators are confronted with. Whenever SELinux is active on the system and the container runtime uses sVirt, you are more likely to need :Z (or :z, as we will see shortly) than not!

Containers are still part of the host operating system. When we create the /srv/db/ postgresql-test location, it will receive the var_t SELinux type by default. Containers that want to use this location would require write privileges to var_t. However, this privilege is not one we want to provide. After all, the containers should be isolated as much as possible from the host—this isolation is what the sVirt technology is about after all. Hence, we need to relabel this location accordingly. The SELinux type to use for generic containers is container_file_t. Moreover, we want to make sure that only the right container can access this location. Restricting and isolating access is what the :Z (with a capitalized Z) does in the command: labeling the directory with the container_file_t type and associating the right categories with it. If we want to have a location accessible by multiple containers, we can tell podman to share the location, yet still be labeled with the container_file_t SELinux type. To accomplish that, we would use the :z argument (with a lowercase z), like so: # podman run -dit --name postgresql-test \ -e POSTGRESQL_PASSWORD="pgsqlpass" \ -v /srv/db/postgresql-test:/bitnami/postgresql:z \ -p 5432:5432 postgresql

310

Enhancing the Security of Containerized Workloads

Creating appropriate mappings is not the only approach where SELinux configuration comes into play. If we want, we can also tell podman to use different SELinux domains for the container as well.

Changing a container's SELinux domain To control the SELinux context under which a container is launched, we use the --security-opt argument to the podman command. For instance, to run an Nginx container with the container_logreader_t SELinux domain, we use the following: # podman run -dit --name nginx-test -p 80:80 \ --security-opt label=type:container_logreader_t nginx

This domain is slightly more privileged than the default container_t domain, as it also has read privileges on log files. We could use this to have a web server expose the log files, for instance. Other labeling options that we can pass on are as follows: • The SELinux user, with the label=user: argument. • The SELinux role, with the label=role: argument. • The SELinux sensitivity level, with the label=level: argument. • The SELinux type for the files, with the label=filetype: argument. This sets the SELinux context for the location mappings that have the :Z and :z suffixes set. The selected type must be an entry point for the container's SELinux domain. There is also another option that we can use, namely label=disable. With this argument set, a container will run without any SELinux isolation. Now, it does not disable SELinux for the container, but associates an unconfined domain called spc_t with the container: # podman run -dit --name nginx-test -p 80:80 \ --security-opt label=disable \ -v /srv/web/localhost:/usr/share/nginx/html nginx # ps -efZ | grep nginx unconfined_u:system_r:spc_t:s0 ... nginx: worker process

While for most use cases, the default container_t domain is sufficiently privileged, it might be too privileged for some. Luckily, we can easily create new SELinux domains specific to our use case.

Configuring podman

311

Creating custom domains with udica The container_t domain is configured to be widely reusable, which implies that it has many privileges for common use cases, which you might not want to give to each container. Furthermore, if we would launch a container but need to associate more privileges with it, then we would have to extend container_t with more privileges, resulting in all containers receiving this privilege extension. To quickly build up new policies, a tool called udica can be used. The udica tool reads the container definition and creates a custom SELinux policy from it. We can then use this custom policy for this particular container, allowing other containers to remain untouched. Let's use this for a Jupyter Notebook, which we want to grant read/write privileges to a (shared) user home directory location: 1. First, we create the definition of the container: # podman run -dit --name notebook-test -p 8888:8888 \ -v /home/lisa/work:/home/jovyan/shared scipy-notebook

2. Next, inspect this container using podman inspect and store the results in a file: # podman inspect notebook-test > notebook-test.json

3. Use udica to generate an SELinux policy for it: # udica -j notebook-test.json custom-notebook-test Policy custom-notebook-test created! Please load these modules using: # semodule -i custom-notebook-test.cil /usr/share/udica/ templates/{base_container.cil,net_container.cil} Restart the container with: "--security-opt label=type:custom-notebook-test.process" parameter

4. Load the custom policy as mentioned by the udica output: # semodule -i custom-notebook-test.cil \ /usr/share/udica/templates/base_container.cil \ /usr/share/udica/templates/net_container.cil

5. Stop and remove the container, and then recreate it with the parameter as mentioned in the udica output: # podman stop notebook-test # podman rm notebook-test

312

Enhancing the Security of Containerized Workloads # podman run -dit --name notebook-test -p 8888:8888 \ -v /home/lisa/work:/home/jovyan/shared \ --security-opt \ label=type:custom-notebook-test.process scipy-notebook

The custom SELinux policy has the privileges to write to the home directory, as the container had a mapping from /home/lisa/work, and udica automatically created the permissions for it. If we wanted the container to only have read-only privileges, we could use a mapping with a trailing :ro (rather than :Z or :z for SELinux-specific changes). This would map the location inside the container with read-only access, and udica would only create read privileges for the associated SELinux type. If creating custom policies is a bit too specific, we can also fine-tune the privileges of the container_t domain with the appropriate SELinux booleans.

Toggling container_t privileges with SELinux booleans The container_t SELinux domain is associated with the svirt_sandbox_domain attribute, and through that association, will automatically be managed by several of the virt_* SELinux booleans that we saw in Chapter 9, Secure Virtualization. There are a few container-specific SELinux booleans as well: • With container_use_cephfs, containers can use CephFS-based storage. This is predominantly used when the containers are managed by larger container-cluster software such as Kubernetes. • With container_manage_cgroup, containers can manage cgroups. This is needed when the container hosts systemd inside, which is often the case for full-blown container runtimes (rather than process-specific containers). Such containers host almost complete Linux systems. • With container_connect_any, the container_t SELinux domain can connect to any TCP port. Keep in mind though that these booleans influence the privileges of the container_t domain, and thus are in effect for all containers.

Tuning the container hosting environment The podman utility will by default store its container volumes and base images in /var/ lib/containers. Administrators can add more locations through the storage. conf configuration file available in /etc/containers. However, you need to adjust the SELinux configuration accordingly as well.

Leveraging Kubernetes' SELinux support

313

Suppose that the /srv/containers location will be used, then we need to create an equivalence rule to make sure that this location is labeled appropriately: # semanage fcontext -a -e /var/lib/containers \ /srv/containers # restorecon -R -v /srv/containers

If the location is a network mount, you might need to change the appropriate SELinux booleans as well.

Leveraging Kubernetes' SELinux support When containers are used in a larger environment, they are often managed through container orchestration frameworks that allow scaling container deployment and management across multiple systems. Kubernetes is a popular container orchestration framework with a good community, as well as commercial support. Kubernetes uses the container software found on the machines under the hood. When, for instance, we install Kubernetes on Fedora's CoreOS, it will detect that Docker is available and use the Docker engine for managing the containers.

Configuring Kubernetes with SELinux support Installing Kubernetes can be a daunting task, and several methods exist, ranging from single-node playground deployments up to commercially supported installations. One of the well-documented installation methods on the Kubernetes website is to use kubeadm for bootstrapping Kubernetes clusters. Important note The installation of Kubernetes is documented on the Kubernetes website at https://kubernetes.io/docs/setup/productionenvironment/tools/kubeadm. In this section, we will not go through the individual steps to set up a working Kubernetes instance, but give pointers as to which changes are needed for having proper SELinux support.

The kubeadm command, when initializing the Kubernetes cluster, will download and run the various Kubernetes services as containers. Unfortunately, Kubernetes' services use several mappings from the host system into the container to facilitate their operations. These mappings are not done using the :Z or :z options—it would even be wrong to do so, as the locations are system-wide locations that should retain their current SELinux labels.

314

Enhancing the Security of Containerized Workloads

As a result, Kubernetes' services will be running with the default container_t SELinux domain (as Docker will happily apply the sVirt protections), which does not have access to these locations. The most obvious change we can apply is to have the services run with the highly privileged spc_t domain for now. Applying this change however during the installation is hard, as we would need to change the domain sufficiently quickly before the installation fails. While we can create deployment configuration information for Kubernetes that immediately configures the services with spc_t, another method can be pursued: 1. Mark the container_t type as a permissive domain before the installation starts. While this will prevent any SELinux controls on the container, we can argue that the installation of Kubernetes is done in a contained and supervised manner: # semanage permissive -a container_t

2. Run kubeadm init, which will install the services on the system: # kubeadm init

3. When the services are installed, go to /etc/kubernetes/manifests. Inside this directory, you will find four manifests, each one representing a Kubernetes service: # cd /etc/kubernetes/manifests

4. Edit each manifest file (etcd.yaml, kube-apiserver.yml, kubecontroller-manager.yml, and kube-scheduler.yml) and add a security context definition that configures the service to run with the spc_t domain. This is done as a configuration directive under the containers section: apiVersion: v1 kind: Pod metadata: name: etcd spec: containers: - command: … securityContext: seLinuxOptions: type: spc_t image: k8s.gcr.io/etcd:3.4.3-0 …

Leveraging Kubernetes' SELinux support

315

5. During the Kubernetes installation, the kubelet service will be installed, which will detect that these files have been changed, and will automatically restart the containers. If not, you can shut down and remove the container definitions within Docker, and kubelet will automatically recreate them: # docker ps CONTAINER ID ... NAMES 548f0c3ed18e k8s_POD_etcd-ppubssa3ed_kube… b7b1df2d0027 k8s_POD_kube-apiserver-… eecd4d4ad108 k8s_POD_kube-scheduler-… 76da4910b927 k8s_POD_kube-controller-… # for n in 548f0c3ed18e b7b1df2d0027 eecd4d4ad108 76da4910b927; do docker stop $n; docker rm $n; done

6. Verify that the services are now running with the privileged spc_t domain: # ps -ef | grep spc_t

7. Remove the permissive state of container_t so that it is back to enforcing mode: # semanage permissive -d container_t

With these slight adjustments during the installation, Kubernetes is now running fine with SELinux support enabled.

Setting SELinux contexts for pods Within Kubernetes, containers are part of pods. A pod is a group of containers that all see the same resources and can interact with each other seamlessly. Previously, in the Configuring podman section, we worked on the container level. The podman utility is also able to use the pods concept (hence the name). For instance, we could put the Nginx container in a pod called webserver like so: # podman pod create -p 80:80 --name webserver # podman pull docker.io/library/nginx # podman run -dit --pod webserver --name nginx-test nginx

Unlike podman, Kubernetes does not rely on command-line interaction to create and manage resources such as pods. Instead, it uses manifest files (as we've briefly touched upon in the Configuring Kubernetes with SELinux support section). Kubernetes administrators or DevOps teams will create manifest files and apply those to the environment.

316

Enhancing the Security of Containerized Workloads

For instance, to have the Nginx containers run on Kubernetes, the following manifest could be used: apiVersion: apps/v1 kind: Deployment metadata: name: nginx-test-deployment spec: selector: matchLabels: app: nginx-test replicas: 2 template: metadata: labels: app: nginx-test spec: containers: - name: nginx-test image: nginx:latest ports: - containerPort: 80

This manifest is a Kubernetes deployment, and tells Kubernetes that we want to run two Nginx containers. To apply this to the environment, use kubectl apply: $ kubectl apply -f simple-nginx.yml

As with the manifests for the Kubernetes services, we can tell Kubernetes to use a specific SELinux type: … spec: containers: ... securityContext: seLinuxOptions: type: "container_logreader_t"

The seLinuxOptions block can contain user, role, type, and level to define the SELinux user, SELinux role, SELinux type and SELinux sensitivity level.

Summary

317

Unlike the regular container management services (such as Docker or CRI-O), Kubernetes does not allow changing SELinux labels on mapped volumes (except on single-node deployments): when we map volumes into containers, they retain their current SELinux label on the system. Hence, if you want to make sure that the resources are accessible from a regular container_t domain, you need to make sure these locations are labeled with container_file_t. Kubernetes does offer advanced access controls itself. Enabling volumes within the containers is also handled by a plugin architecture, with several plugins already available. When the plugin enables SELinux labeling, then Kubernetes will attempt to relabel the resource and assign the categories (as with sVirt). However, this support is currently only made available on single-node deployments (using the local host storage plugin)—and for such deployments, using podman is much simpler.

Summary Containerized workloads allow administrators to add capabilities quickly and easily to a system, while retaining possible dependencies within a container. Each container hosts its own dependencies, allowing containers to be removed and added from the system without affecting others. With SELinux, this workload is further isolated from the host and, in case of sVirt protections, also from each other. We've seen how systemd has container support but lacks sVirt-based protections, and how podman can apply sVirt protections on its own container environments. We learned that Docker and podman are very similar in usage, yet different under the hood. Both frameworks allow us to apply different SELinux types to the containers and resources, and with udica we've learned how to create custom policies without much development effort. Finally, we've seen how Kubernetes can be configured to use SELinux labeling as well. With all these SELinux-capable technologies behind us, we are ready to tackle the SELinux policy development itself. In the next chapter, we'll learn to work with SELinux policies in depth and tune the policies to our needs.

Questions 1. Why is Docker or podman preferred over machinectl for SELinux? 2. How do we ensure host data is properly mapped within a container? 3. How can we create a custom policy from a container definition? 4. Where in Kubernetes' manifests can we place SELinux settings?

Section 3: Policy Management

SELinux behavior is defined by the policy. In this part, we cover policy changes and development, including the creation of custom policies. This section comprises the following chapters: • Chapter 12, Tuning SELinux Policies • Chapter 13, Analyzing Policy Behavior • Chapter 14, Dealing with New Applications • Chapter 15, Using the Reference Policy • Chapter 16, Developing Policies with SELinux CIL

12

Tuning SELinux Policies Until now, we have been working with an existing SELinux policy by tuning our system to deal with the proper SELinux contexts and assigning the right labels to files, directories, and even network ports. We've learned that the behavior that SELinux enforces is defined within the policies. To fine-tune the policy enforcement rules, we have already briefly covered SELinux booleans.

It's time we look into SELinux booleans in more detail, learning how to look up the impact booleans have. Within this chapter, we then consider SELinux policy modules themselves and what options administrators have when dealing with these modules. Finally, we will look at how to update or even replace existing policies. In this chapter, we're going to cover the following main topics: • Working with SELinux booleans • Handling policy modules • Replacing and updating existing policies

322

Tuning SELinux Policies

Technical requirements Check out the following video to see the Code in Action: https://bit.ly/2T7MkVK

Working with SELinux booleans One of the methods of manipulating SELinux policies is by toggling SELinux booleans. Ever since Chapter 2, Understanding SELinux Decisions and Logging, where we used the secure_mode_policyload boolean, these tunable settings have been popping up over the course of this book. With their simple ON/OFF state, they enable or disable parts of the SELinux policy. Policy developers and administrators use SELinux booleans to toggle parts of the policy that not all deployments always need to be active, but some still do. These booleans are added to the policy based on feedback from, and with the help of, the community at large. By establishing which policy rules are necessary against those that are optional, SELinux developers can provide an SELinux policy that works for a majority of systems, even when the uses of these systems differ.

Listing SELinux booleans An overview of SELinux booleans can be obtained by using the semanage command with the boolean option. On a regular system, we can easily find over a hundred SELinux booleans, so it is necessary to filter them out for the description of the boolean we need: # semanage boolean -l | grep policyload secure_mode_policyload (off, off) Boolean to determine whether the system permits loading policy, setting enforcing mode, and changing boolean values. Set this to true and you have to reboot to set it back.

The output not only gives us a brief description of the boolean, but also the current value (actually, it gives us the current value and then the value pending a policy change, but this will almost always be the same). Another method for getting the current value of a boolean is through the getsebool command, as follows: # getsebool secure_mode_policyload secure_mode_policyload --> off

Working with SELinux booleans

323

If the name of the boolean is not exactly known, we can ask for an overview of all booleans (and their values) and filter for the one we need: # getsebool -a | grep policy secure_mode_policyload --> off

Another utility that can be used to view SELinux boolean descriptions is the sepolicy booleans command: # sepolicy booleans -b secure_mode_policyload secure_mode_policyload=_("Boolean to ...")

However, this command does not show the current value of the boolean. Finally, booleans are also represented through the /sys/fs/selinux filesystem: # cat /sys/fs/selinux/booleans/secure_mode_policyload 0 0

Here, booleans can be read as if they were regular files, and they return two values: • The first value is the current state of the boolean, where 0 means OFF and 1 means ON. • The second value is the pending state of the boolean. A pending state allows administrators to change multiple boolean values simultaneously, but only when manipulating booleans through the /sys/fs/selinux filesystem, as we will see next.

Changing boolean values We can change the value of a boolean using the setsebool command. For instance, to toggle the httpd_can_sendmail SELinux boolean, we can use the following command: # setsebool httpd_can_sendmail on

Some Linux distributions might also have the togglesebool command available. This command will flip the value of the boolean, so ON becomes OFF, and OFF becomes ON: # togglesebool httpd_can_sendmail

324

Tuning SELinux Policies

SELinux booleans have a default state defined by the policy administrator (and thus the default SELinux policy active on the system). Changing the value using setsebool updates the current active access controls, but this does not persist across reboots (if we toggle the boolean, then after rebooting, the old value will be used again). In order to keep the changes permanently, add the -P option to the setsebool command as follows: # setsebool -P httpd_can_sendmail off

In the background, the updated SELinux boolean value is included in the policy store. Then, the current policy file is rebuilt and loaded. As a result, the policy file (called policy.## with ## representing an integer value) residing in /etc/selinux/ targeted/policy will be regenerated. This regeneration takes time, which is why switching a boolean value persistently (using -P) takes more time to complete than when we change a value without persisting it (using setsebool without -P or togglesebool) to the policy store. Another way to change and persist the boolean settings is to use the semanage boolean command as follows: # semanage boolean -m --on httpd_can_sendmail

In this case, we modify (-m) the boolean value and set it to ON (--on). Booleans can also be changed through their /sys/fs/selinux/booleans representation. When this happens, the boolean value is not immediately activated – the change of the value is pending. This allows administrators to modify multiple booleans through /sys/fs/selinux/booleans first: # echo 0 > /sys/fs/selinux/booleans/httpd_can_sendmail # getsebool httpd_can_sendmail httpd_can_sendmail --> on pending: off

To commit the changes, write the value 1 into /sys/fs/selinux/commit_ pending_bools: # echo 1 > /sys/fs/selinux/commit_pending_bools

As long as you modify booleans through the semanage or setsebool commands though, the changes will immediately be committed. Only operations through the /sys/fs/selinux structure allow pending boolean changes.

Handling policy modules

325

Inspecting the impact of a boolean To discover which policy rules a boolean manipulates, the description usually suffices. Sometimes though, we might want to know which SELinux rules change when we alter a boolean value. With the sesearch application, we can query the SELinux policy, displaying the rules affected by a given boolean. To show this information in detail, we use the -b option (for the boolean) and -A option (to show all allow rules): # sesearch -b httpd_can_sendmail -A allow httpd_suexec_t bin_t:dir { getattr open search }; [ httpd_can_sendmail ]:True ... allow system_mail_t httpd_t:process sigchld; [ httpd_can_ sendmail ]:True

When we query the SELinux policy directly, conditional rules can be shown as part of the output: # sesearch -s system_mail_t -t httpd_t -A allow domain domain:key { link search }; allow system_mail_t httpd_t:fd use; [ httpd_can_sendmail ]:True ...

When allow rules are suffixed with an SELinux boolean between square brackets followed by :True, then these rules are only applied if the boolean is active. If the boolean is followed by :False, then the rule is applied if the boolean is not active. Not all situations can be perfectly defined by policy writers though. Sometimes we will need to create our own SELinux policy modules and load those. Let's see how we can handle SELinux policy modules specifically.

Handling policy modules When the system loads the SELinux policy in memory, it uses the policy.## file, with ## representing the policy version, as explained at the end of Chapter 1, Fundamental SELinux Concepts. This file, which resides in /etc/selinux/targeted/policy, is generated every time the policy is modified. This can be when booleans are changed (and persisted), or when SELinux policy modules are added or removed.

326

Tuning SELinux Policies

Listing policy modules SELinux policy modules are sets of SELinux rules that can be loaded and unloaded. These modules, with .pp or .cil suffixes, can be loaded and unloaded as needed by the administrator. Once loaded, the policy module is made part of the SELinux policy store, and will be loaded even after a system reboot. Unlike SELinux boolean changes, SELinux policy module loads are always persisted. To list the currently loaded SELinux policy modules, we recommend using the semodule command. By default, semodule will show all loaded SELinux policy modules without any details: # semodule -l abrt accountsd ... zosremote

SELinux policy modules can, however, be loaded at a specified priority. This allows administrators to load a policy that overrules an already loaded policy: SELinux policy modules with a higher policy module priority take precedence over similarly named SELinux policy modules with lower priorities. To see the current priorities, use the --list-modules=full argument: # semodule --list-modules=full 100 abrt pp 100 accountsd pp ... 400 test cil ... 100 zosremote pp

Alongside the priority, the listing also shows whether the policy module is based upon the binary module format (pp) or the more modern Common Intermediate Language (CIL) format (cil). The SELinux utilities will copy the active policy modules into a policy-specific location. This allows administrators to list the currently active modules through regular filesystem queries as well: # ls /var/lib/selinux/targeted/active/modules/* /var/lib/selinux/targeted/active/modules/100: abrt accountsd

Handling policy modules

327

... /var/lib/selinux/targeted/active/modules/400: test

The use of the filesystem location for querying active policies is, however, not recommended, as we have no guarantee that the loaded policies match the filesystem: non-SELinux utilities can add or remove files from these locations without adjusting the SELinux policy state.

Loading and removing policy modules In the Replacing and updating existing policies section, we will learn how to generate new policy modules. Once created, they need to be loaded and/or removed. We load policy modules with semodule as well, regardless of the policy format (.pp or .cil): # semodule -i screen.pp

By default, SELinux policy modules are loaded at the 400 priority when invoked by the administrator, whereas SELinux policy modules loaded as part of the default system policy will be loaded at the 100 priority. When loading policies, the priority can be adjusted using the -X option. For instance, to load the test.cil policy with a priority of 500 we use the -X option as follows: # semodule -X 500 -i test.cil libsemanage.semanage_direct_install_info: Overriding test module at lower priority 400 with module at priority 500.

To remove a policy module with semodule, use the --remove or -r option. In this case, we are not referring to an SELinux policy module file, but to the name of the module itself as displayed by semodule. Hence, we do not need to pass on a suffix: # semodule -r test

To remove an SELinux policy module from a specified priority, use the -X option: # semodule -X 500 -r test libsemanage.semanage_direct_remove_key: test module at priority 400 is now active.

The order of the arguments is important: the -X option will set the priority for the actions that follow it, not those that precede it. If it is not set, then a priority value of 400 will be used.

328

Tuning SELinux Policies

Finally, it is possible to keep an SELinux policy module but disable it. This keeps the module in the policy store, but disables all the SELinux policy rules inside of it. We use the --disable (or -d) option to accomplish this: # semodule -d screen

To re-enable the policy, use the --enable (or -e) option: # semodule -e screen

The disabled and enabled states of SELinux policy modules persist through reboots as well. Furthermore, if you are disabling an SELinux module, all instances of that module (including lower priority ones) will be disabled. Disabling policies is strongly recommended when the policy module is part of the distribution's SELinux policy, as the modules themselves are not always available on the system and might require a reinstallation of the policy package just to get it back. With loading and unloading policies explained, let's see how we can generate updates on the current SELinux policy.

Replacing and updating existing policies When we replace or update existing policies, we need to load them using the semodule commands, as shown in the Handling policy modules section. But how do we create or update the policies, exactly? Let's consider a few use cases where SELinux policy adjustments are triggered.

Creating policies using audit2allow When SELinux prevents certain actions, we know it will log the appropriate denial (assuming no dontaudit statements are defined) in the audit logs. This denial can be used as the source to generate a custom SELinux policy that allows the activity. Consider the following denial, which occurred when a confined user called su to switch to the root user: type=AVC msg=audit(...): avc: denied { write } for pid=58002 comm="su" name="btmp" dev="vda1" ino=4213650 scontext=staff_u:staff_r:staff_t:s0 tcontext=system_u:object_r:faillog_t:s0 tclass=file permissive=0

Replacing and updating existing policies

329

If we are certain that these operations need to be granted, then we can use the audit2allow command to generate a policy module for us that allows these activities. The audit2allow application transforms a denial (or set of denials) into SELinux allow rules. These rules can then be saved in a file, ready to be built into an SELinux policy module, which can then be loaded. To generate SELinux policy allow rules, pipe the denials through the audit2allow application: # grep btmp /var/log/audit/audit.log | audit2allow #============ staff_t ============ allow staff_t faillog_t:file write;

Based on the denials, audit2allow prepared an allow rule. We can also ask audit2allow to immediately create an SELinux policy module: # grep btmp /var/log/audit/audit.log | audit2allow -M localpolicy ************ IMPORTANT ************ To make this policy package active, execute: semodule -i localpolicy.pp

A file called localpolicy.pp will be available in the current directory, which we can load using the semodule command. The source file will also be present, named localpolicy.te. If the denials that occurred are considered cosmetic in nature (meaning that the system functions as expected and the denials should not cause any updates on the policy), you can use audit2allow to generate dontaudit rules rather than allow rules. In that case, the denials will no longer be visible in the audit logs, while still preventing the actions from taking place: # grep btmp /var/log/audit/audit.log | audit2allow -D -M localpolicy

It is likely, after including the necessary rules, that the action will result in more denials that were not previously triggered. As long as the previous AVC denials are still available in the audit logs, it is sufficient to regenerate the policy and continue. After all, audit2allow will consider all AVC denials that it encountered, regardless of the current SELinux policy state.

330

Tuning SELinux Policies

Another popular approach is to put the system (or the application domain) in permissive mode to generate and fill up the audit logs with all the AVC denials related to the action. Although this generates more AVC denials to work with, it could also result in wrong decisions by the audit2allow command. Hence, always verify the denials before generating new policy constructs, and review the generated policy to make sure that it will enforce the right set of access controls and not grant more privileges than needed. When the previous AVC denials are no longer available inside the audit log, a new policy module will need to be generated, as otherwise the previously fixed accesses will be denied again: the newly generated policy will no longer contain the allow rules from before, and when we load the new policy, the old policy is no longer active.

Using sensible module names In the previous section, we used the audit2allow command to generate a policy module named localpolicy. However, this name does not reveal what the purpose of the module is. Once we create a (binary) policy (such as the localpolicy.pp file) and load it, it is not always clear to the administrators and users at first glance what this module is meant to accomplish. Although it is possible to unpack the .pp file (using semodule_ unpackage) and then disassemble the resulting .mod file into a .te file, it requires software not available on most distributions (the dismod application, for instance, part of the checkpolicy software, is often not included). Considering that we just want to get insights into the rules that are part of a module, this is a very elaborate and time-intensive approach. The content of a module can also be somewhat deduced from its CIL code. For instance, an active screen module will have its code available at /var/lib/selinux/ targeted/active/modules/100/screen, in a file called cil. On some distributions, this file will be a compressed file, so you might need to unzip it before viewing: # file screen/cil cil: bzip2 compressed data, block size = 500k # bzcat screen/cil (typealias secadm_screen_home_t ...

Still, having to dive into the rules to know what localpolicy is about is not only very cumbersome, but also requires sufficient privileges to be able to read these files.

Replacing and updating existing policies

331

Instead, it is a best practice to name the generated modules for their intended purposes. An SELinux policy that fixes a few AVC denials that come up when su executes from within the staff_t domain would be best named custom_staff_su_faillog, for instance. It is also recommended to prefix (or suffix) the custom policies, so they can be more easily found: # semodule -l | grep ^custom_ custom_staff_su_faillog

This identifies that the policy module has been added by the administrator (or organization) and is not sourced from the default Linux distribution's policy.

Generating reference policy style modules with audit2allow The reference policy project provides distributions and policy writers with a set of functions that simplify the development of SELinux policies. As an example, let's see what the reference policy functions (called macros) can do with the su situation: # grep btmp /var/log/audit/audit.log | audit2allow -R require { type staff_t; } #============ staff_t ============ auth_rw_faillog(staff_t)

The rule in the example is auth_rw_faillog(staff_t). This is a reference policy macro that explains an SELinux rule (or set of rules) in a more human-readable way. In this case, it allows the staff_t domain to read/write on faillog_t labeled resources. The faillog_t type is part of the system authentication SELinux policy (as suggested by the auth_ prefix, which identifies the source SELinux policy module). Important note As audit2allow -R uses an automated approach for finding potential functions, we need to review the results carefully. Sometimes it selects a method that creates far more privileges for a domain than needed.

332

Tuning SELinux Policies

All major distributions base their SELinux policies upon the macros and content provided by the reference policy. The list of methods we can call while building SELinux policies is available on the local filesystem, at /usr/share/doc/selinux-policy/html. These named methods bundle a set of rules related to the functionality that SELinux policy administrators want to enable. For instance, the storage_read_tape() method allows us to enhance an SELinux policy, providing a given domain with read access to tape storage devices.

Building reference policy - style modules If we generate an SELinux policy using reference policy macros but do not have access to the binary policy module anymore, then we need to build the policy before loading it. CIL-based policies can be loaded directly, which is why this book uses CIL as much as possible. However, given the wide use of the reference policy, knowing how to build these modules is important as well. Suppose that the reference policy-based SELinux policy code resides in a file called custom_staff_su_faillog.te, then we can build it into a .pp file as follows: # make -f /usr/share/selinux/devel/Makefile custom_staff_su_ faillog.pp Compiling targeted custom_staff_su_faillog.pp policy package rm tmp/custom_staff_su_faillog.mod tmp/custom_staff_su_faillog. mod.fc

Once built, we can load it using semodule. Every time we change the policy code (in the .te file) or other policy information (such as file context definitions in the .fc file), we need to rebuild the .pp file before we can load it.

Building legacy-style modules If we ask audit2allow to generate the policy rules without using reference policy style macros (which we call a legacy-style SELinux policy), then building the .pp file from it requires a different approach. Suppose we have the .te file as generated by audit2allow -M, but not the .pp file, then we can generate it as follows: 1. First, create the .mod file using checkmodule: # checkmodule -M -m -o custom_nonrefpol.mod \ custom_nonrefpol.te

Replacing and updating existing policies

333

2. Next, generate the .pp file using semodule_package: # semodule_package -o custom_nonrefpol.pp \ -m custom_nonrefpol.mod

If an .fc file (which contains file context definitions) is present, use the -f option: # semodule_package -o custom_nonrefpol.pp \ -m custom_nonrefpol.mod -f custom_nonrefpol.fc

The audit2allow command will automatically execute these commands, so this is only needed if the .pp file is no longer present, or when these more legacy-style SELinux policies are shared with you and you need to build and load them manually.

Replacing the default distribution policy When adding custom SELinux policies, all that users can do is to add more allow rules. SELinux does not have a deny rule that can be used to remove currently allowed access rules from the active policy. If the current policy is too permissive for the administrator's liking, then the administrator will need to update the policy rather than just enhance it. That implies that the administrator has access to the current SELinux policy rules in use. To replace an active SELinux policy, most Linux distributions allow you to get the source code of the policy. For instance, for RPM-based Linux distributions, the source RPM of the SELinux policy package can be downloaded and unpacked to gain access to the policy as follows: 1. First, find out what the current version of the SELinux policy is: $ rpm -qi selinux-policy Name : selinux-policy Version : 3.14.3 ... Source RPM : selinux-policy-3.14.3-20.el8.src.rpm ...

2. Next, try to obtain the source RPM shown in the output. Source RPMs can also be downloaded from third-party repositories. If the package is difficult to find, you can try to find it through https://rpmfind.net. 3. Next, use the rpmbuild utility to extract the source RPM: $ rpmbuild --rebuild --nobuild \ selinux-policy-3.14.3-20.el8.src.rpm

334

Tuning SELinux Policies

This will unpack the source RPM in the ~/rpmbuild directory. 4. When finished, the SELinux policy source code can be found inside ~/rpmbuild/ SOURCES and is probably named selinux-policy-9c02e99.tar.gz or similar, which you can extract further: $ tar xvf selinux-policy-9c92e99.tar.gz $ tar xvf selinux-policy-contrib-c8ebb9f.tar.gz

The SELinux policy can then be found in the created subdirectories. For instance, the screen.te file can be found in the ./selinux-policy-contribc8ebb*/policy/modules/contrib subdirectory. The policy files can now be safely copied over, manipulated at will, and built to replace the existing policy. If we load the updated SELinux policy module with the same (or higher) priority as the already loaded policy, it will take precedence in the policy. Most distributions will also have their active SELinux policy available through an online source-controlled repository. For instance, the current SELinux policy for CentOS is available at https://github.com/fedora-selinux/selinux-policy.

Summary The SELinux policy can be adjusted by administrators, either through SELinux booleans as provided by the SELinux policy itself, or by loading new SELinux policy modules. These modules can be generated automatically, or built manually by the policy developers. In this chapter, we've learned how to use SELinux booleans and how to query the active policy for the effects that the booleans will have on the system. We then learned how to use semodule to load and unload policies, or enable/disable modules on the system. We ended the chapter with information on how to generate and replace policies. In the next chapter, we will extend our query of the SELinux policy beyond just booleans, and learn how to analyze policy behavior in detail using specialized tools.

Questions 1. How can we mark a boolean change as pending but not commit it yet? 2. Which command can be used to query the impact of a boolean? 3. Why can SELinux policy modules be loaded with different priorities? 4. How can denials be transformed into new SELinux policy modules?

13

Analyzing Policy Behavior Although SELinux policies enforce the mandatory access controls and application behavior on a system, knowing how a policy will act upfront is useful for administrators to perform assessments and root cause analysis activities. Throughout this chapter, we will learn how to query the SELinux policy in depth, using a multitude of tools to query process transitions, analyze information flows, and compare policies. We will consider the apol tool, a graphical interface with which we can perform several analyses on a policy, as well as command-line tools such as sesearch, sedta, seinfoflow, and sepolicy. Finally, we will use sediff to compare policies. In this chapter, we're going to cover the following main topics: • Performing single-step analysis • Investigating domain transitions • Analyzing information flow • Comparing policies

336

Analyzing Policy Behavior

Technical requirements Check out the following video to see the Code in Action: https://bit.ly/3lY56LB

Performing single-step analysis Until now, we've covered a few methods of analyzing SELinux policies through command-line utilities such as seinfo and sesearch. These utilities can assist users in performing single-step analyses: they either provide immediate information about an SELinux object (which is mainly what seinfo is about) or are capable of querying direct SELinux rules (which is the scope of sesearch). Not all capabilities of the seinfo and sesearch utilities have been discussed yet though, so let's see what other tricks these commands have up their sleeves.

Using different SELinux policy files Many SELinux analysis tools, including seinfo and sesearch, can access both the currently loaded SELinux policy and a specified SELinux policy file. The latter allows developers to query SELinux policies of systems they do not have direct access to, for which direct access is cumbersome (such as mobile devices), or that have been used in previous situations (backups) and are no longer active. For instance, to analyze a policy file called policy.sepolicy-2, the following seinfo command can be used: $ seinfo ./policy.sepolicy-2 Statistics for policy file: ./policy.sepolicy-2 Policy Version: 30 (MLS enabled) Target Policy: selinux Handle unknown classes: deny Classes: 63 Permissions: Sensitivities: 1 Categories: Types: 1858 Attributes: Users: 1 Roles: Booleans: 0 Cond. Expr.: Allow: 108120 Neverallow: Auditallow: 24 Dontaudit: Type_trans: 639 Type_change: Type_member: 0 Range_trans: Role allow: 0 Role_trans: Constraints: 0 Validatetrans: MLS Constrain: 59 MLS Val. Tran:

286 1024 28 2 0 0 553 0 0 0 0 0

Performing single-step analysis Permissives: Defaults: Allowxperm: Auditallowxperm: Initial SIDs: Genfscon: Netifcon:

0 0 185 0 27 83 0

Polcap: Typebounds: Neverallowxperm: Dontauditxperm: Fs_use: Portcon: Nodecon:

337

2 0 0 0 16 0 0

When the command is not explicitly told to parse a given policy file, it will try to query the current active policy through the /sys/fs/selinux/policy pseudo-file.

Displaying policy object information The main purpose of the seinfo application is to display SELinux object information. This information is presented on a per-object basis. Various SELinux object types are supported, ranging from the well-known types, attributes, roles, and users, to the more specialized fs_use_* declarations or genfscon statements. A complete list of supported object types (and their resulting seinfo options) can be found on the seinfo manual page, or through the direct help: $ seinfo --help

Regardless of the object type that the user is interested in, seinfo has three main modus operandi: • In the first mode, it lists the objects of a given type. For this, only the option has to be passed on, without additional information. For instance, to list all object classes available in the policy, use the following command: $ seinfo --class

• In the second mode, it can confirm (or deny) the presence of an object instance. To accomplish this, add the instance name to the command. For instance, to validate if the memprotect class is available in the policy, use the following command: $ seinfo --class memprotect Classes: 1 memprotect

Sadly, regardless of whether the instance is available or not, the return code of the application is the same, so scripts cannot use this without additional statements to verify whether the instance exists. Instead, they will need to check the output of the command (which will state that zero instances exist that match the name).

338

Analyzing Policy Behavior

• The third mode displays expanded information about a selected instance. Although not all information objects support an expanded set, most of the common ones do. The expanded information generally shows a list of (different) instances related to the initial query: $ seinfo --class memprotect -x Classes: 1 class memprotect { mmap_zero }

The supported types that seinfo can query are the following: • With --attribute (-a), seinfo shows all currently known SELinux attributes in the policy. When expanded, it shows the types associated with a given attribute. • With --bool (-b), seinfo shows all currently known SELinux booleans in the policy. When expanded, it shows the current value of the boolean. • With --class (-c), seinfo shows the supported SELinux classes. When expanded, it shows the permissions supported by that class. • With --role (-r), seinfo shows the SELinux roles supported in the policy. When expanded, it shows the domains allowed for that role. • With --type (-t), seinfo shows the SELinux types in the policy. When expanded, it shows the aliases that the type has, as well as the attributes. • With --user (-u), seinfo shows the SELinux users (not the Linux users or logins) known by the policy. When expanded, it shows the roles and sensitivity range associated with the SELinux user. • With --category, seinfo shows the currently supported categories. When expanded, it shows the sensitivities for which the category is associated (only in MLS policies). • With --common, seinfo shows the common permission sets. These are sets inherited by different classes. When expanded, it shows the permissions part of that set. • With --constrain, seinfo shows the current constraints. There is no expanded information for this query. • With --default, seinfo shows the default_* rules within the policy. One of these rules, for instance, is the default sensitivity range for a class (default_range). There is no expanded information for this query.

Performing single-step analysis

339

• With --fs_use, seinfo shows the fs_use_* rules within the SELinux policy. One of these rules is to allocate a security context for filesystems that support extended attributes (fs_use_xattr). There is no expanded information for this query. • With --genfscon, seinfo shows the context allocations for filesystems that do not support extended attributes. There is no expanded information for this query. • With --initialsid, seinfo shows all the initial Security Identifiers (SID). These are all the classes that have predefined contexts set. When expanded, it shows the context associated with the SID. • With --netifcon, seinfo shows the contexts currently associated with the network interfaces. This is only applicable when labeled networking is active. There is no expanded information for this query. • With --nodecon, seinfo shows the contexts currently associated with the node definitions (hosts). This is only applicable when labeled networking is active. There is no expanded information for this query. • With --permissive, seinfo shows which types are currently marked as permissive domains. There is no expanded information for this query. • With --polcap, seinfo shows the policy capabilities (that is, in-policy settings that define the SELinux subsystem behavior, such as support for SCTP through the extended_socket_class policy capability that we saw in Chapter 5, Controlling Network Communications). When expanded, it shows the actual policy capability statements in the policy. • With --portcon, seinfo shows the current port mappings and their associated contexts (which is also interpreted by semanage port -l). There is no expanded information for this query. • With --sensitivity, seinfo shows the currently supported sensitivity levels. When expanded, it shows the actual policy statements to declare the sensitivities. • With --typebounds, seinfo shows the type bounds (SELinux types or domains bounded by a parent domain). There is no expanded information for this query. • With --validatetrans, seinfo shows the transition constraints active in the policy (these are constraints that limit when a file transition is allowed). This is not used in most Linux distributions. There is no expanded information for this query. The seinfo command also has an --all option that shows all possible information it can get from a policy file. However, this does not include the expanded information.

340

Analyzing Policy Behavior

Understanding sesearch Where seinfo displays information about SELinux objects, sesearch is used to query SELinux rules and behavior information between a source and a target resource. We have been using sesearch to query standard allow rules (type enforcementrelated access controls) as well as the impact of SELinux booleans on these allow rules. The sesearch application allows us to not just query rules based on the rule type, but also filter based on additional parameters. Let's see which parameters can be used for sesearch filters: • The most common queries are to filter out the rules that match a given source expression using --source (-s) and/or target expression using --target (-t): $ sesearch -A -s mount_t -t unconfined_t

The sesearch application can also deal with indirect source or target information. For instance, when querying information related to the svirt_sandbox_domain attribute, it will also display rules of all types that have this attribute assigned. We can selectively disable this behavior using -ds (for source) and -dt (for target): $ sesearch -A -s svirt_sandbox_domain -ds

• With the --class (-c) argument, we can search for only those rules affecting a specified resource class (such as file, dir, tcp_socket, and so forth—the list of all possible classes can be obtained using seinfo --class): $ sesearch -A -s svirt_sandbox_domain -c file

• If we are interested in only a particular action (or permission), we can use the --perm (-p) argument. This is particularly useful when we encounter a denial for a certain action (say, write) and want to see which domains are allowed to perform this action, as it might indicate that we are examining the wrong source domain. We can list multiple permissions, in which case sesearch will display the rules that have at least one permission in them: $ sesearch -A -s staff_t -c file -p write

With the -ep option, sesearch will only list the rules that have all permissions in them, rather than at least one. • We can also query only those rules influenced by an SELinux boolean using the --bool (-b) argument, as we saw in Chapter 12, Tuning SELinux Policies. If we use the -eb option, then all booleans listed on the command line must be matched, rather than at least one.

Performing single-step analysis

341

• The sesearch application can also use regular expressions rather than actual values. This is not the default behavior, but can be activated with -rs (for the source type or role), -rt (for the target type or role), -rc (for the class), -rd (for the default type or role), and -rb (for the boolean): $ sesearch -A -s staff_.*_t -c process -p transition -rs

As this provides insights into the most common SELinux behavior and access controls, let's go through the various rules and the impact they have on the system.

Querying allow rules The first set of rules are the allow rules, of which many subtypes exist. The standard allow rule defines which actions a source domain can successfully trigger toward or against a target type: $ sesearch --allow -s guest_t -t cgroup_t -c dir allow guest_usertype cgroup_t:dir { getattr ioctl lock open read search }; allow guest_usertype filesystem_type:dir { getattr open search };

There are a few similar rules that SELinux policies can define, and these can be queried similarly with sesearch as follows: • Using --auditallow, we can query which actions are allowed by SELinux but will still result in an audit event. • Using --dontaudit, we can query which actions will not trigger an audit event, even when the action is denied. • Using --neverallow, we can query which actions are forbidden from being declared within the policy. Such actions, when defined, will cause the system to refuse to load new SELinux policies if they violate the rule. It cannot be used to negate existing rules though, and neverallow rules that you attempt to add to the policy afterward will fail if the current policy already has deviations against this rule. SELinux also supports extended permission rules. These rules are similar to regular allow rules but take an additional (number) parameter that further limits the applicability of the rule, and are used to provide fine-grained access control for device queries. These queries are generally handled by the ioctl() system call, but until its support for extended permissions, SELinux could only control whether a domain was allowed to use the ioctl() system call or not, rather than filtering on the explicit query through ioctl().

342

Analyzing Policy Behavior

With extended permission rules, SELinux policy developers can specify which ioctl() queries are allowed and which ones aren't. For instance, we can grant a domain the ability to get a hardware address (known as SIOCGIFHWADDR, which is defined with number 0x8927) as follows: allowxperm : tcp_socket ioctl 0x8927;

Within sesearch, we can query these rules using --allowxperm. Like regular allow rules, we also have the --auditallowxperm, --dontauditxperm, and --neverallowxperm options to cover the extended-permission-equivalent rules. These have the same impact on the query and also on the extended permission rules.

Querying type transition rules A second set of rules are the type transition rules. Rather than informing the system which actions are allowed or not, type transitions influence the SELinux context of objects and resources through actions taken by the processes on the system. Type transition rules, for instance, define what context a newly written file receives when it is written by a particular domain within a directory that has a given SELinux type, or what domain a newly created process receives when it is executed from a given source domain: $ sesearch -T -s guest_t -c process type_transition guest_t abrt_helper_exec_t:process abrt_ helper_t; type_transition guest_t chfn_exec_t:process chfn_t; ...

In this output, we can see that when the guest domain successfully executes a binary labeled with abrt_helper_exec_t, its resulting process will be assigned the abrt_helper_t context. These rules are queried and interpreted by various tools in the Investigating domain transitions section.

Performing single-step analysis

343

Querying other type rules Alongside the allow rules and type transition rules, sesearch can also query two other type-related rules: type_change and type_member. These rules are meant for SELinux-aware applications and are not enforced by the in-kernel SELinux subsystem: • With type_change statements (which can be filtered in sesearch using the --type_change option), developers inform the SELinux-aware application that a target resource should be relabeled with a given type on behalf of the source domain. For instance, when systemd assigns a terminal to a user, it queries the SELinux policy for type_change statements for the user domain, given the current terminal's SELinux type, and will return the following type_change statement: type_change guest_t tty_device_t:chr_file user_tty_ device_t;

As the device file itself already exists and is only reassigned to a user, no type transition itself is done. Instead, the type_change rule is interpreted by the SELinux-aware application that relabels the device file accordingly. • The type_member rule (which can be filtered in sesearch using the --type_ member option) informs SELinux-aware applications that participate in setting up polyinstantiated locations (as we saw in Chapter 3, Managing User Logins) about the target SELinux type of such directories. For instance, when the /tmp location (which is labeled with tmp_t) is polyinstantiated for a user, then the following rule is used to understand that the /tmp view for this user is to be labeled with user_tmp_t: type_member guest_t tmp_t:dir user_tmp_t;

The PAM module responsible for addressing the polyinstantiation is SELinux-aware, and will use these rules to deduce what the target types must be for the created locations. Alongside the type-related statements, sesearch can also handle role-related queries.

344

Analyzing Policy Behavior

Querying role-related rules SELinux also has rules related to role activities and transitions. With the sesearch application, we can query which SELinux roles are allowed to be accessed from other roles, and when a role transition (such as switching from a user role to the system role) is performed: $ sesearch --role_allow allow dbadm_r sysadm_r; $ sesearch --role_trans role_transition dbadm_r role_transition dbadm_r system_r;

-s dbadm_r; -s dbadm_r; mysqld_initrc_exec_t:process system_r; postgresql_initrc_exec_t:process

The distinction between the two is that the allowed access (using --role_allow) shows which roles can be accessed from a given role, but they do not dictate when the change is done. The role transitions (using --role_trans) show when the system attempts to automatically change a role (and to what role it would be changed) when executing a script or binary. Hence, they can be compared with the allow rules (which specify what is allowed) and type transitions (defining SELinux behavior). Analyzing role transitions and role allow rules helps administrators deduce which roles are too powerful or could result in potential security issues. For instance, allowing the dbadm_r role to switch to the system_r role through the postgresql_initrc_ exec_t type might allow that role to invoke actions outside its scope if it also has the rights to modify postgresql_initrc_exec_t resources: $ sesearch -A -s dbadm_t -t postgresql_initrc_exec_t -c file; allow dbadm_t postgresql_initrc_exec_t:file { execute execute_ no_trans getattr ioctl map open read };

While directly modifying postgresql_initrc_exec_t files is thus not allowed, it is not enough to only look at the main user type. A decent analysis needs to include all types reachable by the dbadm_r role, which we will cover in the Investigating domain transitions and Analyzing information flow sections. These sections will use apol, so let's first see what this application is all about.

Performing single-step analysis

345

Browsing with apol An advanced tool to perform policy analysis is apol, which can be launched by just executing the command without any arguments. The tool is graphical in nature and allows analysts and administrators to perform a wide range of analytical actions against the SELinux policy. Once started, the first action to take with apol is to load a target policy (either the currently active policy or a file copied over from a different system). This can be accomplished through the Open Policy button, or by navigating to File | Open Policy. The tool will then display a generic overview of the loaded policy:

Figure 13.1 – The apol application after loading a policy file

346

Analyzing Policy Behavior

Once it has been loaded, select New Analysis to initiate the policy analysis functions:

Figure 13.2 – apol's overview of supported analysis methods

Performing single-step analysis

347

A decent number of analysis methods are provided. Let's select Types to browse through the available types, or select an attribute to find out which SELinux types are assigned to that attribute:

Figure 13.3 – Type browsing with apol

348

Analyzing Policy Behavior

Similarly, with the TE Rules analysis, we can perform the same analysis as we did with the sesearch application:

Figure 13.4 – Querying type enforcement rules with apol

The more advanced analysis methods are covered in the Investigating domain transitions and Analyzing information flow sections.

Using apol workspaces Analyzing SELinux policies can take a while, especially when this involves multiple phases of analysis and fine-tuning. The apol tool allows you to save your current workspace to disk, so that you can later get back to the analysis from the point at which you saved it:

Investigating domain transitions

349

Figure 13.5 – Workspace management in apol

Workspaces not only retain the settings of the queries so far, but also the notes you might add. Notes are an important feature within apol where you can write down thoughts and observations from the queries you've made. The notes are associated with the tabs you have open, allowing you to switch between different queries as needed. Now that we know how the apol application works, let's see how we can use it (and other tools) for more in-depth analyses.

Investigating domain transitions An important analytical approach when dealing with SELinux policies is to perform a domain transition analysis. Domains are bounded by the access controls in place for a given domain, but users or processes can transition to other domains by executing the right set of applications.

350

Analyzing Policy Behavior

Analyzing whether and how a transition can occur between two SELinux domains allows administrators to validate the secure state of the policy. Given the mandatory nature of SELinux, adversaries will find it difficult to be able to execute target applications if a domain transition analysis shows that the source domain cannot execute said application, either directly or indirectly. Administrators should use domain transition analysis to confirm a domain is correctly confined, and that vulnerabilities within the applications running inside a domain cannot lead to privilege escalations.

Using apol for domain transition analysis After starting apol, select New Analysis followed by Domain Transition Analysis. The analysis screen itself will show several possible analytical approaches:

Figure 13.6 – Querying possible transition paths between staff_t and unconfined_t

Investigating domain transitions

351

This analysis will attempt to find a path between a given source domain and target domain, and display the execution trail that could lead to the transition. Administrators can then verify whether the applications associated with these domain transitions can be trusted or not. Such analysis is sensible when we need to assert that certain domains cannot break out of their confinement, or when we are developing new policies and want to ensure that the confinement is within the boundaries we want. The transition analysis can be fine-tuned through the following settings: • With Shortest paths, apol will show domain transitions between the source domain and the target domain, seeking the shortest transitions possible. For instance, a transition from staff_t to staff_sudo_t to unconfined_t is a two-step path. When a path is found, apol will not search for longer paths. • When we select All paths up to, apol will perform the analysis up to a certain number of steps. When we use up to one step, then this is similar to doing direct queries with seinfo or sesearch. • Using Transitions out of the source domain and Transitions into the target domain will show all transitions that can occur from a given source domain or to the target domain. This is used for a more interactive session, where users can click through the domains to see the next set of domains that can be transitioned to. To further fine-tune the analysis, a few options can be selected. For instance, we can exclude certain types from being used in the domain transition analysis. This allows us to mark certain domains as trusted (such as the *_sudo_t domains), which will make apol ignore those domains to find more appropriate transition chains to analyze.

Using sedta for domain transition analysis The path analysis done by apol can also be executed from a command-line application called sedta. It has the same capabilities as the domain transition analysis functionality within apol. The type of analysis is selected through command-line arguments: -S is used for shortest path analysis, whereas -A (followed by a number) runs the equivalent of All paths up to. For instance, to check for a domain transition path between the staff_t domain and the unconfined_t domain, excluding the staff_sudo_t, newrole_t, and init_t domains, use the following command: $ sedta -S -s staff_t -t unconfined_t staff_sudo_t newrole_t Domain transition path 1: Step 1: staff_t -> oddjob_t

352

Analyzing Policy Behavior

Domain transition rule(s): allow staff_t oddjob_t:process transition; Set execution context rule(s): allow staff_t staff_t:process { dyntransition fork getattr getcap getpgid getrlimit getsched getsession noatsecure rlimitinh setcap setcurrent setexec setfscreate setkeycreate setpgid setrlimit setsched setsockcreate share sigchld siginh sigkill signal signull sigstop transition }; Entrypoint oddjob_exec_t: Domain entrypoint rule(s): allow oddjob_t oddjob_exec_t:file { entrypoint execute getattr ioctl lock map open read }; File execute rule(s): allow staff_t oddjob_exec_t:file { execute execute_no_ trans getattr ioctl map open read }; Type transition rule(s): type_transition staff_t oddjob_exec_t:process oddjob_t; Step 2: oddjob_t -> openshift_initrc_t ...

We can analyze a different policy than the current system policy using the -p option.

Using sepolicy for domain transition analysis The sepolicy tool has a built-in domain transition analysis capability using the transition argument. It is, however, not as flexible as sedta or apol, as no tuning can be done to the command. It also does not seem to cover all possible paths, often displaying extensive and elaborate routes that could be much simpler: $ sepolicy transition -s mount_t -t unconfined_t mount_t ... glusterd_t ... ipsec_t ... ipsec_mgmt_t ... initrc_t ... condor_schedd_t ... condor_startd_t ... openshift_initrc_t ... stunnel_t ... telnetd_t ... remote_login_t @ shell_exec_t --> unconfined_t -- Allowed True [ unconfined_login=1 ] mount_t ... glusterd_t ... ipsec_t ... ipsec_mgmt_t ... initrc_t ... condor_schedd_t ... condor_startd_t ... openshift_initrc_t ... kmscon_t ...

Investigating domain transitions

353

local_login_t @ shell_exec_t --> unconfined_t -- Allowed True [ unconfined_login=1 ] mount_t ... glusterd_t ... ipsec_t ... ipsec_mgmt_t ... initrc_t ... condor_schedd_t ... condor_startd_t ... openshift_initrc_t ... kdumpgui_t ... kdumpctl_t ... sge_execd_t ... sge_shepherd_t ... sshd_t @ shell_exec_t --> unconfined_t -- Allowed True [ ssh_sysadm_login=0 || unconfined_login=1 ] mount_t ... glusterd_t ... ipsec_t ... ipsec_mgmt_t ... initrc_t ... condor_schedd_t ... condor_startd_t ... openshift_initrc_t ... kdumpgui_t ... kdumpctl_t ... sulogin_t @ shell_exec_t --> unconfined_t -- Allowed True [ unconfined_login=1 ] mount_t ... glusterd_t ... ipsec_t ... ipsec_mgmt_t ... initrc_t ... condor_schedd_t ... condor_startd_t ... openshift_initrc_t ... kdumpgui_t ... kdumpctl_t ... inetd_t ... rshd_t @ shell_exec_t --> unconfined_t -- Allowed True [ unconfined_login=1 ] mount_t ... glusterd_t ... ipsec_t ... ipsec_mgmt_t ... initrc_t ... condor_schedd_t ... condor_startd_t ... openshift_initrc_t ... kdumpgui_t ... kdumpctl_t ... piranha_pulse_t ... crond_t @ shell_exec_t --> unconfined_t -- Allowed True [ cron_userdomain_transition=1 || unconfined_ login=1 ] mount_t ... glusterd_t ... ipsec_t ... ipsec_mgmt_t ... initrc_t ... condor_schedd_t ... condor_startd_t ... openshift_initrc_t ... kdumpgui_t ... kdumpctl_t ... piranha_pulse_t ... cockpit_ws_t ... cockpit_session_t @ unconfined_exec_t --> unconfined_t

Let's compare this with sedta, which we use against the same policy and for the same domain transition: $ sedta -S -s mount_t -t unconfined_t | \ grep -E '(transition path|Step)' Domain transition path 1: Step 1: mount_t -> glusterd_t Step 2: glusterd_t -> sulogin_t Step 3: sulogin_t -> unconfined_t Domain transition path 2:

354

Analyzing Policy Behavior

Step 1: mount_t -> glusterd_t Step 2: glusterd_t -> virtd_lxc_t Step 3: virtd_lxc_t -> unconfined_t Domain transition path 3: Step 1: mount_t -> glusterd_t Step 2: glusterd_t -> xdm_t Step 3: xdm_t -> unconfined_t Domain transition path 4: Step 1: mount_t -> glusterd_t Step 2: glusterd_t -> crond_t Step 3: crond_t -> unconfined_t Domain transition path 5: Step 1: mount_t -> glusterd_t Step 2: glusterd_t -> sshd_t Step 3: sshd_t -> unconfined_t Domain transition path 6: Step 1: mount_t -> glusterd_t Step 2: glusterd_t -> virtd_t Step 3: virtd_t -> unconfined_t 6 domain transition path(s) found.

When comparing the transition paths with the ones generated by sedta, you will notice that sedta often finds shorter domain transitions, which sepolicy transition does not. Hence it is not recommended to rely solely on sepolicy transition for domain transition analysis.

Analyzing information flow Another analytical investigation that can be carried out on SELinux policies is information flow analysis. Unlike domain transitions, which look at how one domain can gain a certain set of permissions through transitions to another domain, information flow analysis looks at how a domain could leak (purposefully or not) information to another domain. Information flow analysis is performed by looking at all operations that occur between two types. A source type can be read by a domain, which subsequently can write information to another type that can then be accessed by another domain. While this can still be analyzed in a step-wise fashion, it quickly becomes very challenging because we cannot limit ourselves to the read and write operations. Information can be leaked through filenames, file descriptors, and more. Information flow analysis must take all these methods into account.

Analyzing information flow

355

Using apol for information flow analysis After loading an SELinux policy, select Information Flow Analysis. The interface we receive is similar to the domain transition analysis, but now has a few toggles to fine-tune the path analysis specific to information flows:

Figure 13.7 – Analyzing information flow between two domains

Unlike domain transitions, the number of paths through which information can flow is exponentially bigger. To perform a decent information flow analysis, we need to fine-tune the search criteria: • The Minimum permission weight option allows users to only look at permissions or actions that have a particular weight. Each action is given a weight in the tool, from a low priority (such as the lock operation, which has weight 1) to a high priority one (such as the write operation, which has weight 10). The purpose of these weights is to define which actions are plausible for information flow and which ones are much harder (but not impossible) to use for deliberate information exchange.

356

Analyzing Policy Behavior

• With Excluded Permissions, we can selectively enable or disable certain permissions from the analysis. The other options are similar to those in domain transition analysis. The most important area for information flow analysis is the permission map, which we can fine-tune partially while enabling or disabling permissions in the analysis. However, we might not be happy with the weights that the current permission map uses. To edit the permission map, select Permission Map | Edit Permission Map from the apol menu:

Figure 13.8 – Editing the permission map and permission weights

Within this editor, we can fine-tune the weights of the permissions to our liking, as well as the directionality of the action: • None (no information flow) • Write (information flows to the resource)

Analyzing information flow

357

• Read (information is retrieved from the resource) • Both (information can both flow to and from the resource) Once we are satisfied with the results, we can (and probably should) save the permission map for later reuse (if not, the changes are only applicable to the current session and will be forgotten when apol is closed).

Using seinfoflow for information flow analysis Like the sedta application for domain transition analysis, there is also a commandline application that offers information flow analysis capabilities similar to apol, that is, seinfoflow. Every invocation of the seinfoflow command requires the permission map to be passed on for its analysis. If you don't have a permission map created and saved yourself, you can use the default one available at /var/lib/sepolgen/perm_map. Let's analyze the information flow possibilities between the staff_t and guest_t domains, using the default permission map, and only considering the permissions of weight 10: $ seinfoflow -S -m /var/lib/sepolgen/perm_map \ -s staff_t -t guest_t -w 10

The more elaborate a permission map is, the more time it takes for the analysis to complete.

Using sepolicy communicate for simple information flow analysis The sepolicy command can perform a simple flow analysis using the communicate option. Given a source and target domain, sepolicy will check through which intermediate types information can flow between the domains: $ sepolicy communicate -s postgresql_t -t staff_t krb5_host_rcache_t cluster_conf_t security_t postgresql_t postgresql_tmp_t hugetlbfs_t

The preceding flow analysis is basically checking what types can be written to by the source domain, and read by the target domain.

358

Analyzing Policy Behavior

Comparing policies Until now, we've analyzed a single policy set, finding the domain transitions and information flow paths. The commands and applications we've used all focus on this single-policy analysis. Another important analysis is to compare two policies. Policy developers can use this to compare a new policy with an old one, or to compare two system policies to see what additional rules have been added by the administrator.

Using sediff to compare policies The sediff tool looks at the differences between two policy files and reports those to the user. It is often not sensible to use this against completely different policies, but is powerful for finding slight differences between policies, which can assist in troubleshooting issues across different systems. A common use case for sediff is to validate that a source-built policy file is the same as the distribution-provided binary policy file. Administrators can then be certain that the source code they've used to build a policy file is the same as that used by the distribution, even when the binary files themselves (the policy.## file) have different checksums: $ sediff policy.31 /sys/fs/selinux/policy Policy Properties (0 Modified) Classes (1 Added, 0 Removed, 4 Modified) Added Classes: 1 + xdp_socket Modified Classes: 4 * capability2 (1 Removed permissions) - compromise_kernel * process (1 Added permissions, 1 Removed permissions) + getrlimit ...

It is possible to direct sediff to only show differences for a specified area or part of the SELinux policy (such as the available types, roles, booleans, or type enforcement rules).

Summary

359

Summary SELinux has quite a few analysis tools that we can use to analyze policies. We've seen how to use sesearch to do in-depth assessments of the current policy, but noticed that it fails to validate the more dynamic analysis requirements. With apol, we have seen a graphical application that is able to do more dynamic analysis, including the domain transitions (examining which domains can be reached from a current point) and information flow analysis (investigating how information can flow from one domain to another). From this experience, we've learned that such analysis is intensive and requires lots of interpretation to be done correctly. Next to apol, we also learned that command-line utilities exist with similar capabilities: sedta for domain transition analysis, seinfoflow for information flow analysis, and sepolicy, which has a few out-of-the-box functionalities, but not as extensive or flexible as the other options we looked at. In the end, we learned how to compare policies using sediff. This is useful for when new policies are being developed, which is something we'll do in the remaining chapters. We first start with aligning and extending existing policies for new applications in the next chapter, and move on to full application policy development in the last two.

Questions 1. What is the difference between seinfo and sesearch? 2. How do you check whether you can reach a domain? 3. Why does analyzing information flows take so long? 4. Can we generate a delta between policies and load it?

14

Dealing with New Applications New applications are often not yet supported through an application-specific SELinux policy, as most application projects do not develop the SELinux policies themselves, but rely on the community in general (or Linux distributions more specifically) to create and maintain them. Some Linux distributions have implemented fallbacks to allow these applications to run, even though they might not be isolated properly. Administrators might not like the sound of having untrusted new applications running without any SELinux enforcements active though. Hence, this chapter covers how administrators can run new applications in a number of isolated environments, ranging from the (often default) unprotected domains, to sandbox systems, and eventually by reusing existing SELinux domains without having to develop completely new ones. In this chapter, we're going to cover the following main topics: • Running applications without restrictions • Using sandboxed applications • Assigning common policies to new applications • Extending generated policies

362

Dealing with New Applications

Technical requirements Check out the following video to see the Code in Action: https://bit.ly/3dGG5Bu

Running applications without restrictions The default behavior in many Linux distributions is to run new applications through unconfined domains. These are specially crafted domains that, while still being controlled by SELinux, are designed to have very, very broad permissions granted. You can compare such unconfined domains with a firewall that allows any possible flow: while the firewall is running, it is hardly doing any enforcement. There is, however, another approach possible as well, namely, running an application as a permissive domain. Unlike unconfined domains, permissive domains are not enforced through SELinux: everything the domain does is allowed, even though SELinux might log every violation. We briefly touched upon permissive domains in Chapter 3, Understanding SELinux Decisions and Logging. Let's first look at unconfined domains and how administrators can modify system configuration to apply unconfined domains to other applications, or remove applications from being unconfined.

Understanding how unconfined domains work An unconfined domain is an SELinux domain that has broad permissions, restricting only a very small amount of actions that a domain can do. Unconfined domains are not really a concept that SELinux, as technology, supports. Instead, it is used by SELinux policy developers who created a set of permissions they consider as being unconfined. End users on many Linux distributions will have noticed that their own context is unconfined_t. While this is indeed a reference to being an unconfined domain, there are more domains that are unconfined than unconfined_t. SELinux policy developers have aggregated most of the permissions related to unconfined domains either in the domains themselves (as is the case for the reference policy) or in SELinux attributes, such as unconfined_domain_type and unconfined_user_ type (as is the case for CentOS and related Linux distributions). In the case of attributes, these attributes are then assigned to one or more domains to effectively make them unconfined in nature: $ seinfo -a | grep unconfined $ seinfo -a unconfined_domain_type -x

Running applications without restrictions

363

Once a process is running as an unconfined domain, that does not imply that every action of that domain remains unconfined. When an unconfined domain executes a process that has a proper SELinux policy assigned, it is possible for this execution to still invoke a domain transition, effectively running the executed command in a (possibly confined) SELinux domain. As the decision whether a domain transition is allowed or not falls within the SELinux policy, it is recommended that administrators query which domain transitions are allowed and which ones aren't. We saw how to analyze domain transitions in Chapter 13, Analyzing Policy Behavior. Given that we are mostly interested in single-step analysis, we can use the sesearch utility to have a quick overview of supported domain transitions: $ sesearch -A -s unconfined_service_t -c process -p transition allow unconfined_service_t chronyc_t:process transition; allow unconfined_service_t rpm_script_t:process transition; allow unconfined_service_t unconfined_service_t:process { transition ...}; allow unconfined_service_t virt_domain:process { transition ...};

We can see the (many) permissions related to an unconfined domain by either checking them for a single domain, or for the attribute that represents unconfined domains directly: $ sesearch -A -s unconfined_domain_type -ds

Using unconfined domains is preferred over making domains permissive, so let's see how we can mark a new application to run as an unconfined domain.

Making new applications run as an unconfined domain When applications are executed, there are a number of checks that need to pass before this results in a domain transition: • The source SELinux domain must be able to execute the application (implying execute rights on the SELinux type associated with the application's binary or script). • The source SELinux domain must be able to transition to the target domain. • The target domain must have its application binary or script labeled with an SELinux type that is marked as an entrypoint for that domain. • The target domain must be allowed for the SELinux role that the source domain is running with (or a role transition has to be allowed, but that is a corner case).

364

Dealing with New Applications

All these checks are related to the SELinux policy and the labels. It comes as no surprise then that, in order for us to enable applications to run in an unconfined domain, we need to associate the right labels. Let's consider two examples in the following sections, one being a user-triggered application, while the other is a daemonized service.

Running applications in an explicit unconfined domain For applications that users execute, let's take the example of Jailkit, which we introduced in Chapter 7, Configuring Application-Specific SELinux Controls. By default, this application is not associated with any domain, so it runs within the same domain as the parent process. If we are logged in to the system through the unconfined_u user (in the unconfined_t SELinux domain), then we have nothing to do. But suppose that our staff user is confined, yet we want to have the command run in the unconfined_t domain. Important note This is used as an example that shows how to have applications run in a target domain – in our case, an unconfined domain. Allowing confined users to run unconfined applications always has a risk associated with it, because they might use this to break out of their confinement. Make sure that this is only done for applications or users where you have confidence that they will not breach security.

To allow the application to run in the unconfined_t domain, we will use sudo and its SELinux support. While we could also extend the SELinux policy to allow it transparently, this is not recommended. Updating the SELinux policy to allow confined users to run unconfined commands implies that several principles listed in the policy are overturned. You would need to allow the confined user to switch to the unconfined_r role (which is often not allowed for security reasons) transparently, for instance. It would require significant analysis to make sure that it cannot be used to break out of the confined role. Using sudo allows us to limit the methods through which such more privileged commands are executed. SELinux-wise, the appropriate controls are put on the staff_sudo_t domain, for instance, which is only assigned when executing the sudo command, rather than the staff_t domain, which is where most of the user's interactions are executed.

Running applications without restrictions

365

Let's allow the lisa user to run the jk_init command as an unconfined process: 1. First, check whether the SELinux user for which we want to execute the command is allowed to do anything with the unconfined_r SELinux role (and if not, add the role to the SELinux user configuration): # semanage user -l

Allowing a role does not imply that the user domain automatically switches role when needed though, but rather that it is an allowed role for the user. 2. Next, update the /etc/sudoers file to include a transition when executing the following command: # visudo lisa ALL=(root) ROLE=unconfined_r TYPE=unconfined_r NOPASSWD: /usr/sbin/jk_init

In this case, we not only use a ROLE and TYPE transition, but we also allow the command to be executed as the root user, as that is a requirement for the jk_init command. Of course, this can be adjusted as needed. 3. Our user can now run the command, prefixed by sudo, to have it execute in the right domain and using the right role: $ sudo /usr/sbin/jk_init -v -j /srv/chroot \ extshellplusnet

Using sudo for end user applications is common when the privileges of the user also have to switch (from the user privilege to the root privilege). It is less common to use it when staying within the Linux user context though.

Running daemons in an explicit unconfined domain The second use case, and perhaps a more common one than for end user applications, is to run daemonized services in an unconfined domain. Most Linux distributions that use unconfined domains (such as CentOS) will by default have newly installed software run as an unconfined domain as well. For instance, any service that is enabled and activated through systemd (which runs as the init_t SELinux domain) and that does not have an explicit labeling set (meaning the executable commands are labeled as bin_t) will run in the unconfined_service_t domain.

366

Dealing with New Applications

But what if we have a confined application that we want to run in an unconfined domain? Let's take PostgreSQL as an example. Suppose this is an isolated database that has certain extensions active that are incompatible with the existing PostgreSQL SELinux domain (postgresql_t). Administrators might not have the time to extend the current SELinux policy using methods such as audit2allow, as seen in Chapter 12, Tuning SELinux Policies. Luckily, we can easily move PostgreSQL to work and run in an unconfined domain. There are two ways to approach this: • We can remove the existing labels on its executable files (postgresql_exec_t) and set it to bin_t instead. This will then trigger the default transition when starting the PostgreSQL binary to the unconfined_service_t domain. • We can update the SELinux policy for postgresql_t to become an unconfined domain itself. Switching the labels is easy, but is the least recommended method. It is, however, a quick and dirty way to see whether running the service in the unconfined_service_t domain is sufficient to resolve the issue immediately: # chcon -t bin_t /usr/bin/postgres

If agreeable, make sure that the label change remains, even after a relabel operation occurs: # semanage fcontext -a -t bin_t /usr/bin/postgres

Updating the SELinux policy for the PostgreSQL daemon is recommended though, as it retains the existing support within the policy (including the file transitions and other integrations that the postgresql_t domain has with other domains and resources). It also allows administrators to update the policy as needed later on, when there is more time available. To make the postgresql_t domain unconfined, we need to assign the unconfined_domain_type attribute to the postgresql_t domain. This can be accomplished by loading in the following CIL-based SELinux policy: (typeattributeset cil_gen_require postgresql_t) (typeattributeset cil_gen_require unconfined_domain_type) (typeattributeset unconfined_domain_type (postgresql_t))

Running applications without restrictions

367

Save this in a file and load it using semodule -i, and from that point onward the postgresql_t domain will be augmented with the privileges associated with the unconfined_domain_type attribute.

Extending unconfined domains As unconfined domains are still enforced, it might be possible that SELinux is still preventing some actions from occurring. We can adjust the SELinux policy to extend unconfined domains with more privileges though. While the default unconfined_ service_t domain has almost all possible permissions set, more specifically, identified domains might not be as expansive. The trick to adding more privileges to the domains is to assign the appropriate attribute to them. The method is the same as seen in Running daemons in an explicit unconfined domain, adding more attributes as needed. The list of attributes that we can add is very significant (as you can see from seinfo -a), but the most important ones, especially for the CentOS-based SELinux policy, are the following: • files_unconfined_type allows the domain to manage any possible file- or filesystem-based resource. • devices_unconfined_type allows the domain to interact and manage any device resource. • filesystem_unconfined_type allows the domain to interact and manage all filesystems. • selinux_unconfined_type allows the domain to interact with and manage the SELinux subsystem and configuration. • storage_unconfined_type allows the domain to interact with storage systems and removable devices. • dbusd_unconfined allows the domain to interact with all possible D-Bus services. • xserver_unconfined_type allows the domain to interact with and manage all X server resources. Furthermore, there are several can_* attributes that fine-tune very specific, security-sensitive actions. The names of these attributes nicely explain what they allow. For instance, can_write_shadow_passwords allows the domain to write to /etc/ shadow, whereas can_change_object_identity means that the domain can change the SELinux user of an object.

368

Dealing with New Applications

Not all attributes have their privileges reflected in regular allow rules or transitions that can be queried using sesearch. For instance, can_change_object_identity is used in SELinux constraints instead: # seinfo --constrain | grep can_change_object_identity

Querying the constraints is an often forgotten method to see what or why a certain privilege is or isn't assigned to a domain. Suppose now that an application still fails to run correctly within an unconfined domain, then we can use permissive domains to allow this application to run unprotected, while having the rest of the system remain in enforcing mode.

Marking domains as permissive As we saw in Chapter 2, Understanding SELinux Decisions and Logging, we can mark a domain as permissive using semanage permissive: # semanage permissive -a postgresql_t

The same command can be used to query (-l) or remove (-d) permissive states. However, administrators should take special care before marking domains as permissive: • First of all, if you mark a domain as permissive, then all processes running with that SELinux domain will run without any active SELinux enforcements. As an administrator, you really want to limit the number of processes that are running through permissive domains, so do not mark broadly used SELinux domains as permissive. A daemon that runs in an unconfined domain, yet still has problems, should not result in the unconfined domain being marked as permissive. Instead, have the daemon run as a different domain, and mark that domain as permissive. • Secondly, permissive domains will still trigger SELinux behavior by the SELinux subsystem. Transition rules, including process transitions and file transitions, are still executed. This is of course by design, as permissive domains are meant to be short-lived, allowing administrators and developers to capture information and adapt the policy as needed before they can remove the permissive flag again.

Using sandboxed applications

369

This also implies that, if the domain does not have proper transition rules set, it might result in files being created on the system that have the wrong SELinux types set. Because of this, using permissive domains should not be considered for applications or daemons that have a wide impact on the system, but rather for more isolated situations where you, as an administrator, feel confident that you can easily fine-tune the policy if needed. Consider the situation where we deploy pgpool-II, a load balancer for PostgreSQL databases, and find that the application does not run properly in an unconfined domain, even though it already runs in the unconfined_service_t SELinux domain. While we can put this domain in permissive mode, this would also apply to various other services running inside the unconfined_service_t domain. What we can do is relabel its resources (executables mostly) so that the application is run through a different SELinux domain, and then mark that domain as permissive. We can either reuse an existing, unused domain or generate one, as we will see in the Generating policies with sepolicy generate section. When we want to run an application in a (strictly) confined manner though, we need to take a completely different route and seek out how to put such applications in sandbox-like domains.

Using sandboxed applications New applications that should only have very limited privileges, and that are untrusted by nature, should be confined completely. While we could look at custom SELinux policies for these applications, this is hardly possible for each and every application out there. Instead, we can consider sandboxing the applications, isolating their access from the system. With the help of some other Linux primitives such as namespace support, a utility has been created called the SELinux sandbox, which launches applications in a tightly confined domain. This is mostly meant for end user applications. Important note The SELinux sandbox, its SELinux policy, and the command associated with it, is specific to Linux distributions that use or follow Red Hat packages, such as CentOS. It might not be available for your Linux distribution.

For service-oriented domains, using the container runtime and protection measures are more suited. For more information about using container protections, see Chapter 11, Enhancing the Security of Containerized Workloads.

370

Dealing with New Applications

Understanding the SELinux sandbox The SELinux sandbox is a combination of a number of technologies and protection measures. While the SELinux policy plays an important part, other isolation measures are taken as well to really create a sandbox experience for applications and users. The purpose of the sandbox is to create a low-privilege environment that blocks anything that could jeopardize the security of the system or the user's data. This also means that network interaction is blocked by default (no data exfiltration), and many system resources are hidden away from the sandboxed process. Many of the access controls themselves are handled by the SELinux policy. The sandbox SELinux domains, sandbox_t, and derivatives such as sandbox_xserver_t, do not have many privileges for other resources. The sandbox utility will also apply sVirt-like categories to differentiate one sandboxed process from another. The isolation, however, is done using different means. Namespaces are used to give the sandboxed process a different view of the filesystem (similar to polyinstantiation), whereas runtime capabilities are dropped before executing the process. The seunshare application is responsible for doing these isolation tasks. Let's see how the SELinux sandbox works in practice.

Using the sandbox command The SELinux sandbox uses the sandbox command. Now, before we can use it, we need to make sure that our SELinux user has multiple categories set as, otherwise, the SELinux sandbox cannot randomly allocate two categories for isolation: # semanage login -l # semanage login -m -r "s0-s0:c0.c100" lisa

Once assigned, we can prepare for running an untrusted application in a sandbox. For instance, we can download one of the International Obfuscated C Code Contest applications from https://www.ioccc.org, compile it, and then only run it in a sandbox mode just in case the code behaves maliciously: 1. Assuming we use the 2019 entry from adamovsky, we should have the prog binary and the advent.unl file ready to use. Create a location in which to store these files, and copy them over: $ mkdir sandbox $ cp 2019/adamovsky/* sandbox

Using sandboxed applications

371

2. Next, run the prog command from within the sandbox: $ sandbox -H sandbox/ prog advent.unl Welcome to Adventure!! Would you like instructions? **

3. While the application runs, we can check its current context with ps: # ps -efZ | grep prog

Alongside the prog command itself, which will be running in the sandbox_t SELinux domain and with a certain category pair set, you will notice that a seunshare command will run alongside it. This command provides the isolation for the process, not only by triggering the SELinux context change, but also removing unnecessary mount and filesystem views from the process's viewpoint. 4. If we exit the application, we can see that the sandbox location has been labeled with an sVirt-like MCS pair: $ ls -Z sandbox/ staff_u:object_r:sandbox_file_t:s0:c29,c94 advent.unl staff_u:object_r:sandbox_file_t:s0:c29,c94 prog

The method we used here was to explicitly tell the sandbox to create an isolated home directory based upon the sandbox/ folder and run the prog binary from within this location (and with advent.unl as an argument to the prog command). However, this is not the sole approach. If no explicit home directory is provided, then the sandbox will create a temporary one (and clean it up afterward). However, in that case, we cannot execute commands that are not already installed on the system, unless we allow the sandbox domain to execute user_home_t-labeled resources: (typeattributeset cil_gen_require sandbox_t) (allow sandbox_t user_home_t (file (execute map)))

With this policy loaded, we can use the sandbox with the least number of options. For instance, with the Burton contest submission (also from IOCCC's 2019 contest), we have the following: $ cat prog.c | sandbox ./prog 1 1 127

372

Dealing with New Applications

The use of a more known location, however, allows more flexibility, as well as allowing the sandbox to keep data across multiple sessions (as the directory pointed toward will not be cleaned up). The SELinux sandbox also supports running graphical applications in the sandbox. To accomplish this, add the -X option to the sandbox command. The resulting process will run in the sandbox_xserver_t domain rather than the sandbox_t domain, as more privileges are needed to allow graphical applications to run. Keep in mind though that the sandbox domain has very few privileges; connecting to networked resources is not allowed, so it is not possible to use the sandbox (without additional modifications and SELinux policy adjustments) to run a sandboxed browser to interact with unsafe websites.

Assigning common policies to new applications In between the strong isolation of an SELinux sandbox and the broad permissions of unconfined domains (or even permissive domains) sits the sufficiently privileged application domain. For most administrators, having a proper SELinux domain for applications is the best way forward, as it allows all the common behaviors and restricts unwanted ones. When we start looking at application domains, however, we notice that there is differentiation in complexity, and as an administrator, we need to understand what the complexity is about before we can make the right choice.

Understanding domain complexity SELinux is able to provide full system confinement: each and every application runs in its own restricted environment that it cannot break out of. But that requires fine-grained policies that are developed as quickly as the new releases of all the applications they confine. Developing fine-grained policies at this speed is not possible, so a balance has to be struck between the maintainability of a policy and the security of the domain. This balance is the policy design complexity or domain complexity, which can be roughly categorized as follows: • Fine-grained policies have separate, individual domains for each sub component of an application or service. Such policies have the advantage that they really attempt to restrict applications as much as possible. Through fine-grained policies, roles developed with users and administrators in mind become fine-grained as well, for instance, by differentiating sub-roles in the application.

Assigning common policies to new applications

373

The disadvantage of such policies is that they are hard to maintain, requiring frequent updates as the application itself evolves. The policies also need to take into account the impact of the various configuration options that the application supports. Such fine-grained policies are not frequently found. An example is the policy set provided for the Postfix mail infrastructure. Each sub-service of the Postfix infrastructure has its own SELinux domain. • Application-level policies use a single domain for an application, regardless of its sub-components. This balances the requirement for application confinement versus the maintainability of the application and its SELinux policy. Such application-level policies are the most common in SELinux policies. They do still suffer from regular maintenance as applications expand their functionality, but the complexity of this is limited and SELinux policy developers should not have too many problems maintaining these policies. • Category-wide policies use a single domain definition for a set of applications that implement the same functionality. This is popular for services that act very similarly and whose user-role definitions can be described without really considering the application-specific nature. A good example of a category-wide policy is the policy for web servers. While this policy was initially written for the Apache HTTP daemon, the policy has become reusable for a number of web servers, such as the Cherokee, Hiawatha, Nginx, and Lighttpd projects. While such policies are easier to maintain, the downside of category-wide policies is that they often have more broad privileges than really needed. As more applications are joined in the category-wide policy, additional rules and privileges are added to support those specific functions. • Coarse-grained policies are used for applications or services whose behavior is hard to define. End user domains are examples of coarse-grained policies, as are unconfined domains. When we are dealing with a new application, and we want to quickly assign a decent-enough policy, the most common method is to see whether a category-wide policy exists that we can reuse for the application.

374

Dealing with New Applications

Running applications in a specific policy Let's consider the situation for the pgpool-II application. When we install it without any additional changes, it will run with the unconfined_service_t domain, as mentioned in the Marking domains as permissive section. But perhaps we can find a suitable policy to run the pgpool-II application with, through which it is more confined. As the pgpool-II solution is a load balance-like application for PostgreSQL databases, it is likely we can run it in the PostgreSQL domain. If there are no PostgreSQL databases running on the same system, then lending this domain for the pgpool-II application might not do much harm. Let's see how well this goes: 1. The PostgreSQL policy uses the postgresql_exec_t SELinux type for its executables, so let's assign that one to the pgpool binary: # chcon -t postgresql_exec_t /usr/bin/pgpool

2. If we try to start the pgpool system service, we might get one or more failures: # systemctl start pgpool # systemctl status pgpool ... WARNING: Failed to open status file at: "/var/log/pgpool/ pgpool_status" FATAL: could not read pid file

3. One of the failures mentioned is that the daemon cannot access its logs (in /var/ log/pgpool) while another complains about the process ID file (in /var/run/ pgpool) being unreachable. As these were previously created by an unconfined domain, it is indeed likely that their context is wrong as well. Let's apply the PostgreSQL-specific types to these locations: # chcon -R -t postgresql_log_t /var/log/pgpool # chcon -R -t postgresql_var_run_t /var/run/pgpool

4. After restarting pgpool, we notice it has a new failure: # systemctl start pgpool # systemctl status pgpool ... LOG: Setting up socket for ::1.9999 FATAL: failed to create INET domain socket DETAIL: bind on socket failed with error "Permission denied"

Extending generated policies

375

This time, we get a permission failure, which most of the time implies that the SELinux policy is refusing a certain action: # ausearch -i -m avc -ts recent ... avc: denied { name_bind } for pid=20065 comm=pgpool src=9999 scontext=system_u:system_r:postgresql_t:s0 tcontext=system_u:object_r:jboss_management_port_t:s0 tclass=tcp_socket

The denial seems to reflect the information displayed earlier: pgpool wants to listen on port 9999, but SELinux is refusing this. 5. Let's create a small policy enhancement to allow postgresql_t to bind to this port: (typeattributeset cil_gen_require jboss_management_ port_t) (typeattributeset cil_gen_require postgresql_t) (allow postgresql_t jboss_management_port_t (tcp_socket (name_bind)))

6. Load this policy and restart pgpool. With this in place, pgpool starts up fine. Of course, having the daemon launch without problems does not mean that it will work without problems, so it is recommended to continue testing, using the service as intended. Finding out which policy can be reused for a process requires a bit of practice and searching. For instance, you can query the policy for which domains are able to bind to the port that the daemon needs. Or you can search for a domain that has a behavior very similar to the application involved. In our example, we only had to allow the domain to bind to port 9999. We could also use this information point to seek a different policy—one that is allowed to bind to this port (such as the httpd_t domain) and see whether that one fits better. While this approach is trial and error, it could allow running the service in a more confined domain than the unconfined domain would. A much better approach, however, is to generate a new, custom policy and work from there.

Extending generated policies When we assign a different policy to a new application, we are reusing and possibly extending existing policies. We can go a step further and generate new policies, after which we can further extend those policies, effectively moving into the realm of developing new policies ourselves.

376

Dealing with New Applications

In Chapter 15, Using the Reference Policy, and Chapter 16, Developing Policies with SELinux CIL, we will expand further into the policy development aspects for more fine-grained control. By using policy generation tools, however, we can quickly create a first-draft policy and adapt as needed. An important caveat is that policy generation tools often limit themselves to a single-policy format, either being reference policy style or CIL style. Administrators and organizations should try to focus on a single style and stick with that so that the learning curve for new developers and administrators isn't too high.

Understanding the limitations of generated policies Policy generators, such as the udica tool we saw in Chapter 11, Enhancing the Security of Containerized Workloads, often have a very specific purpose. For instance, the udica tool focuses on generating new container SELinux domains and is only useful for those containers. Generators will always have a specific target in mind for what their policies should look like. The generated policies are often application-level policies. Creating fine-grained policies with generators is hard, and defining category-wide policies requires multiple steps and occurrences, whereas generators often use single-step generations. Furthermore, most generated policies only generally support role-based access controls within SELinux: either a user is allowed the target SELinux domain and interacting with it, or the user isn't allowed. Differentiating roles (such as application administrator versus application user) are not often included in generated policies. Administrators should be aware that generators also have to make assumptions about how applications work. While this allows generators to be used for the majority of simple services and applications, they are definitely not ready yet to substitute a knowledgeable team of SELinux policy developers.

Introducing sepolicy generate The sepolicy command is able to generate initial SELinux policy modules, which administrators and developers can then fine-tune further. This generator will use some resources on the system (such as the package database of the distribution) to better understand which resources to include, and generates a number of SELinux policy files.

Extending generated policies

377

As there are different types of applications around, the sepolicy generate command also requires the user to inform it about the application type. The following types are currently supported: • User applications are identified with the --application option. Such applications are meant for end users to launch and interact with. • System service applications are identified with the --init option. Applications that run in daemon mode or with their own user are most often system service applications. • D-Bus system service applications are identified with the --dbus option. This type of service is invoked by D-Bus. • Common Gateway Interface (CGI) scripts or applications are supported through the --cgi option. Using CGI-specific domains allows having CGI applications run in their own domain, rather than extending the privileges of the web server domain itself. • Internet services daemon (inetd) applications are supported through the --inetd option. • Sandbox applications are like user applications but much more confined, and are supported through the --sandbox option. Next to application-level policy generation, sepolicy generate also supports generating user domains and roles: • Standard users with support for the graphical desktop can be generated using the --desktop_user option. This is a common, non-administration-oriented user role. • A more lightweight, minimal user role that still supports the graphical desktop can be generated using the --x_user option. This domain focuses on minimal permissions and thus requires further extensions before they can be better put to use. • If no graphical user interface needs to be supported, then you can use the --term_ user option. This generates a confined user domain without desktop support. • Administration-oriented user domains can be generated using the --admin_user option. This is meant for broad administrative privileges. • More confined administration domains can be generated using the --confined_ admin option. This allows you to generate user domains that have administrative roles for a limited number of application domains, not to the system as a whole.

378

Dealing with New Applications

The generator also supports customizing existing domains further (using --customize) or generating specific types (using --newtype). Let's use sepolicy generate to generate a policy for the pgpool-II application.

Generating policies with sepolicy generate The sepolicy generate command will create a skeleton SELinux policy, using the reference policy code style. This policy can then be gradually extended with the privileges the application needs. Let's create and adapt the policy for pgool: 1. First, we tell sepolicy to generate a new policy, named pgpool, which is intended for the /usr/bin/pgpool binary: # sepolicy generate -n pgpool --init /usr/bin/pgpool

2. Next, build the generated SELinux policy: # make -f /usr/share/selinux/devel/Makefile pgpool.pp

3. Load the policy in memory: # semodule -i pgpool.pp

4. Relabel the filesystem, or at least the locations mentioned in the generated pgpool.fc file: # restorecon -RvF /usr/bin/pgpool /var/log/pgpool \ /var/run/pgpool

5. Start the pgpool service: # systemctl start pgpool

After starting, be amazed that pgpool is running flawlessly. Now, you might have the impression that this was too easy. Yes, it was. The default SELinux policy that sepolicy generate provides is permissive, as you can see from within the pgpool.te file: permissive pgpool_t;

Summary

379

If we remove this statement, rebuild, and reload the policy, then we will notice the failures coming up again, such as the process not being allowed to bind to the selected ports. We can now use audit2allow, for instance, to help us extend the policy as needed: # cat /var/log/audit/audit.log | audit2allow -R

Gradually extend, rebuild, and reload the policy until the application works without problems.

Summary Linux administrators can use SELinux controls to prevent or confine access to applications, but this is not always the requirement at hand. Being able to run the application with the right set of permissions is, and what the right set is depends on the user's intentions and the environment. Within this chapter, we've learned how to apply the appropriate confinement to application domains, ranging from very isolated container environments over regular application domains, category-wide permission sets, and up to unconfined domains and even permissive domains. We learned that this is done by first finding the appropriate domain, understanding which labels the domain uses, and then assigning the right labels to the files so that the application is executed in the right domain. We also learned how to generate new policies (using sepolicy generate) ourselves without immediately having to dive into a full SELinux policy development approach, which is what we will consider in the final two chapters.

Questions 1. What is the difference between an unconfined domain and a permissive domain? 2. How can we run applications in a very restricted domain? 3. How can we easily switch the domain in which a service will run? 4. Why do policies generated by sepolicy seemingly run without problems?

15

Using the Reference Policy Up until now, we've covered how to interact with the SELinux subsystem and gradually adjusted the SELinux policy to our liking. As we add more applications and users, we notice that developing custom SELinux policies might help us tune the system more to our liking. There are two main approaches to develop SELinux policies, and using reference policy style development is one of them. The other is discussed in Chapter 16, SELinux Common Intermediate Language. To properly develop an SELinux policy, we'll learn how to use and understand the macros that the reference policy provides, and apply the main coding and development style patterns that the project requires to ensure consistency across SELinux policy modules. We then apply this to two main types of modules: application policies and user policies. In this chapter, we're going to cover the following main topics: • Introducing the reference policy • Using and understanding the policy macros • Creating application-level policies • Getting help with supporting tools

382

Using the Reference Policy

Technical requirements The code files for this chapter can be found in our Git repository at https://github. com/PacktPublishing/SELinux-System-Administration-ThirdEdition. Check out the following video to see the Code in Action: https://bit.ly/3jcBDvI

Introducing the reference policy The reference policy, available through https://github.com/SELinuxProject/ refpolicy, is the source SELinux policy for most, if not all, Linux distributions out there. While it is possible that the plain reference policy will not work out of the box for any Linux distribution (as many Linux distributions add their own touch to the policy, or adjust it so it fits the applications and support tooling installed), the development methodology, structure, and approach used by the reference policy are applicable to all major distribution policies. We recommend checking out the SELinux policy of your distribution to see and easily modify SELinux policies for the system. In this chapter, we'll use a checkout of the reference policy: $ git clone https://github.com/SELinuxProject/refpolicy.git

The SELinux policy repositories for the Linux distributions should be documented by the distributions themselves. A few example repositories are listed next: • For CentOS, the policy repository can be found at https://github.com/ fedora-selinux/selinux-policy. • For Gentoo Linux, the policy repository can be found at https://gitweb. gentoo.org/proj/hardened-refpolicy.git/. • For Debian, the policy repository can be found at https://salsa.debian. org/cgzones/selinux-policy-debian. • For Arch Linux, the policy repository can be found at https://github.com/ archlinuxhardened/selinux-policy-arch/. If the Linux distribution does not have a publicly reachable repository for its SELinux policy, we can often still obtain it through the packages themselves, as used in Chapter 12, Working with SELinux Policies.

Introducing the reference policy

383

While it is not the intention to do full policy rebuilds, we can easily copy over the necessary policy files to our own development environment and fine-tune or extend the policy as needed.

Navigating the policy At its base directory, the reference policy hosts all the common files for building the policies, explaining how to install them, and so forth. The policy itself is in the policy folder, which contains three directories: • flask contains the initial definitions used to jumpstart SELinux, such as listing the supported classes, creating initial security identifiers, and more. We will not touch this location further. • modules contains the SELinux policy code and is the main location for all policy rules. • support contains macros and definitions that are reused across the policy and are not associated with a single policy module. If we enter the modules directory further, we get directories that represent the type of modules or policies contained. This representation by itself is merely to have some structure across the hundreds of modules that are developed: • admin contains system administration related policy modules. • apps contains general application policy modules. • kernel contains core system policy modules (not just kernel related ones). • roles contains SELinux role definitions and default user domain policy modules. • services contains general service policy modules (and is by far the largest set of policy modules). • system contains common system related policy modules. The interpretation of which folder a policy is placed in is left to the reference policy project itself, and discussed on its mailing list when it is not obvious. As policy files are required to have a unique name, we can find the appropriate location easily. For instance, to see where the ipsec policy module is stored: $ ls policy/modules/*/ipsec.te policy/modules/system/ipsec.te

384

Using the Reference Policy

While browsing, you'll notice that the policy modules are always represented by three files, which we describe next.

Structuring policy modules If we analyze an SELinux policy module's code, such as for the dhcp module in the services folder, we'll notice that it has three files associated with it: • dhcp.te, which contains the type enforcement rules, and is the main area of attention for most changes • dhcp.fc, which contains the file context definitions, informing the policy which files or resources need to be labeled with dhcp related SELinux types • dhcp.if, which contains interface definitions, which are reusable functions or macros that can be used in the dhcp SELinux policy code as well as elsewhere Let's quickly see how each of these files is structured.

Understanding type enforcement files The type enforcement file, dhcp.te in our example, has the following structure: policy_module(dhcp, 1.18.2) ## Declarations # SELinux booleans # SELinux types ## Local policy # Internal SELinux rules # Core interfaced SELinux rules # SELinux boolean controlled SELinux rules # Non-blocking interfaced SELinux rules

Let's look at each of these areas with an example. Declaring SELinux objects The Declarations section in a policy tells us what SELinux types, or other SELinux objects such as SELinux booleans and SELinux roles, are defined within this module.

Introducing the reference policy

385

The following declarations are common in SELinux policies: • The first declaration in the dhcp SELinux policy declares an SELinux boolean for this module. It is best practice to start the boolean with the SELinux policy module name, although in this case the choice is made to explicitly use dhcpd rather than dhcp to make it obvious for administrators it is about the DHCP daemon and not possible clients or other use cases: ## ##

Determine whether DHCP daemon can use LDAP ## backends

## gen_tunable(dhcpd_use_ldap, false)

The SELinux boolean is accompanied by a specifically structured comment. Comments within the reference policy that use a double hash prefix (##) will be parsed by the build code and used to update information outside of the SELinux policy. In this case, the description of the SELinux boolean is created, which will be made visible later on through commands such as semanage boolean. Once a module is loaded that defines an SELinux policy, other modules can use this boolean as well. • Some domains might also declare a role attribute, which allows easy management of which roles are allowed to use the domain: attribute_role dhcpd_roles;

• The rest of the declarations in the dhcp SELinux policy declare the SELinux types that the policy owns: type dhcpd_t; # The SELinux domain for the daemon type dhcpd_exec_t; # The executable label for the daemon init_daemon_domain(dhcpd_t, dhcpd_exec_t) # Linking the executable to the domain

SELinux policy modules in the reference policy only declare the types and other objects that they own, not those they use. The objects used but defined by other modules should always be hidden away and interacted with through the interface calls. While other definitions can be added to the section as well, these are the most common. Next up are the local policy rules.

386

Using the Reference Policy

Adding the domain's local rules The local policy within the type enforcement defines the allowed behavior of the domains owned by the SELinux policy module. For the dhcp SELinux policy module, this is only focusing on the dhcpd_t SELinux domain. Other SELinux policy modules, especially if they offer a more fine-grained policy structure, will do this for several of its own SELinux domains, or even SELinux user roles. Let's go through the SELinux policy rules for the dhcp.te example we are looking at: • The policy starts with the internal SELinux rules, which are interactions between the SELinux types owned by the SELinux policy module itself: allow dhcpcd_t self:process { getcap signal_perms }; manage_files_pattern(dhcpd_t, dhcpd_tmp_t, dhcpd_tmp_t)

The most simple rules are the standard allow rules, similar to those that audit2allow would recommend. These allow rules can refer to support macros (such as signal_perms), which we'll discuss in the Using and understanding the policy macros section. The second line, which is a call to manage_files_ pattern, is also a support macro. • The second set of local policy definitions are the core interfaced SELinux rules: kernel_read_system_state(dhcpd_t)

These calls use the code that another SELinux policy module has defined in its interface file. In the case of the kernel_read_system_state interface, this will grant the dhcpd_t SELinux domain the rights to read proc_t labeled resources. As proc_t is not defined by the dhcp SELinux policy module, an interface call has to be used. Core interfaced SELinux rules are rules that should at all times be available to the system. Unlike application related SELinux policy modules, which can be disabled or unloaded, these core rules are associated with type definitions that cannot be removed from the system or disabled at will. • The third set of local policy definitions are the SELinux boolean controlled calls: tunable_policy(`dhcpd_use_ldap', ` # If boolean is true sysnet_use_ldap(dhcpd_t) ', ` # If boolean is false ')

Introducing the reference policy

387

Here, the SELinux interface calls (which can also be standard rules such as allow rules) are surrounded by a tunable_policy() call, which identifies the SELinux boolean (in our case dhcpd_use_ldap) that will influence the SELinux policy rules. Most policy modules will only have a single block (for the rules that are activated if the SELinux boolean is true) but it is possible to have two blocks, where the second one defines the rules in case the SELinux boolean is false. • The final set of local policy definitions are the non-blocking interfaced SELinux rules: optional_policy(` bind_read_dnssec_keys(dhcpd_t) ')

These are the calls that use definitions provided by other SELinux policy modules, but where these SELinux policy modules might not be loaded on the system. In our example, the bind_read_dnssec_keys() call allows the dhcpd_t SELinux domain to read dnssec_t labeled resources, as defined by the bind SELinux policy module. However, BIND might not be installed on the system, and the Linux distribution might thus not have its policy loaded. So this call is optional and only active if the bind SELinux policy module is loaded. The type enforcement file is the file that will change most often. The file context definition file, which we discuss next, is a close second.

Declaring file contexts The file context definition file, with the .fc suffix, tells the SELinux subsystem what SELinux types have to be associated with the file resources on the system. This information is used by tools such as restorecon to reset the context appropriately. The rules inside the file are generally grouped based on the directory for which the rules apply. Each rule is structured like so:

[]

Let's see what these entries imply: • The path expressions are the same ones we saw in Chapter 4, Using File Contexts and Process Domains. Keep in mind that characters such as dot (.) have a specific meaning (in this case, it reflects any possible character) so that paths that really require a dot inside must escape the dot.

388

Using the Reference Policy

• The type/class is an optional setting. If omitted, then it means any possible class is used. The most common values to use are regular files (--), directories (-d), sockets (-s), and symbolic links (-l). • The context is a reference to the target SELinux type for this resource. In the reference policy, these context references always need to be encased by the gen_context() macro, which will add or remove the sensitivity depending on the MLS or MCS support built inside the policy. Let's look at a simple example from the dhcp SELinux policy module: /var/named/data(/.*)? cache_t,s0)

gen_context(system_u:object_r:named_

In this example, the /var/named/data directory, and any resource below it, will be labeled with the named_cache_t SELinux type. The last file that is associated with an SELinux policy module is the interface definition file.

Exposing SELinux rules through interfaces Interfaces within an SELinux policy module are meant to support a more flexible, modular development of SELinux policies across different modules. Whenever a domain or SELinux role needs to interact with resources that are defined in a different SELinux policy module, that module should create a properly named interface for the interaction. Interfaces should be accompanied by a minimal amount of documentation, although this documentation is only used when building the documentation of the entire policy. When this is done, the resulting documentation is made available on the system, for instance in /usr/share/doc/selinux-policy/html. Let's look at the definition for the dhcpd_domtrans() interface: ######################################## ## ## Execute a domain transition to run dhcpd. ## ## ## ## Domain allowed to transition. ## ## # interface(`dhcpd_domtrans',` gen_require(`

Introducing the reference policy

389

type dhcpd_t, dhcpd_exec_t; ') corecmd_search_bin($1) domtrans_pattern($1, dhcpd_exec_t, dhcpd_t) ')

As is best practice, the interface name starts with the SELinux policy name and is followed by the action that is allowed. Sometimes, this action is suffixed with the target resource. The interface itself can reference the arguments passed along to the interface using $1 (first argument), $2 (second argument), and so on. So a call such as dhcpd_ domtrans(init_t) will have the interface called, where $1 is substituted with init_t. Let's look at a few examples with common actions: • Domain transitions allow a domain to transition to another domain. If the SELinux policy module only manages a single domain, then there is no target resource defined (as with dhcpd_domtrans()). If there are multiple domains, then the target resource will define which domain (as with bind_domtrans_ndc()) to allow a domain transition to the ndc_t SELinux domain. • Permission interactions generally refer to the permission at hand and, if it is possible to misinterpret to which resource the permission refers, the target resource is listed. So dhcpd_setattr_state_files() allows the domain to set the attributes of the dhcpd_state_t labeled resources, whereas bind_signal() allows the domain to send signals to the named_t labeled processes. Most of the interface definitions will be permission interactions. • Role-oriented interfaces will grant the associated SELinux role and SELinux domain all the privileges needed to perform a functional role for the SELinux domain. For instance, the dhcpd_admin() role will allow managing the dhcpd related resources, starting and stopping the dhcpd service, and so on. Suppose you want to grant this to the web administration role, then the call within the web administration SELinux policy module could look like so: optional_policy(` dhcpd_admin(webadm_r, webadm_t) ')

390

Using the Reference Policy

While developing SELinux policies, it is recommended to look into the interface definitions of the SELinux policy modules to see which ones exist and what they provide. Policy developers will put often requested permissions in such interfaces, so the available interfaces give a good view of what you will most likely need for your own SELinux policy module. The interface definitions are also made available on the system at /usr/share/ selinux/devel/include so that you can create and modify SELinux policy modules even without checking out the main source repository. Whenever we build a reference policy style module, we use a command like the following: $ make -f /usr/share/selinux/devel/Makefile .pp

This will cause the build process to look for the interfaces inside the /usr/share/ selinux/devel/include location, as well as inside the current working directory.

Using and understanding the policy macros Across the various SELinux policy definitions, we have come across macros that are not tied to a specific SELinux policy module. These are support macros, available inside the policy/support/*.spt files. The most common macros are those declared inside the obj_perm_sets.spt file (which group common permissions for the same class in a single definition) and the *_patterns.spt files (which group permissions across different classes in a single definition).

Making use of single-class permission groups Single-class permission groups allow developers to ignore possible extensions of the SELinux supported permissions as time goes by. For instance, if you want to allow a domain to execute a certain resource, it is most often not enough to allow the execute permission. You also need the open and read permissions (as otherwise, the domain cannot read the executable) and the map permission (to allow mapping the file in memory). If you were to put all these permissions in your own SELinux policy module, then the rule could look like so: allow dhcpd_t dhcpd_exec_t:file { getattr open map read execute ioctl execute_no_trans };

Using and understanding the policy macros

391

If, later on, the SELinux policy is extended with an additional permission that is associated with executing resources, then you will need to look for and update these permissions all over the different SELinux policy modules. So the reference policy moves all these permissions in a macro called exec_file_ perms, defined as follows: define(`exec_file_perms',`{ getattr open map read execute ioctl execute_no_trans }')

With this macro defined, our policy line can be simplified as follows: allow dhcpd_t dhcpd_exec_t:file { exec_file_perms };

If at any point the permissions need to be extended, all that has to happen is to extend the macro definition itself, and the SELinux policy modules can be left untouched.

Calling permission groups While single-class permission groups are a good use for simplifying policy development, permission groups that cover multiple classes are even more common. For instance, if a domain needs full management privileges (implying read, write, as well as creating and removing resources) on resources inside /var/lib/dhcpd, then not only are these privileges needed on the files inside that directory (which are labeled with the dhcpd_state_t SELinux type), but you also need read/write permissions on the directory itself. Such a privilege definition would result in something like so: allow $1 dhcpd_state_t:dir { rw_dir_perms }; allow $1 dhcpd_state_t:file { manage_file_perms };

Rather than declaring these as separate calls, they can be put into a single one that groups the two: manage_files_pattern($1, dhcpd_state_t, dhcpd_state_t)

SELinux policy developers best get acquainted with the various macros available to allow for rapid and efficient SELinux policy development.

392

Using the Reference Policy

Creating application-level policies Application-level policies provide confinement for applications or services. There are a number of different types of application-level policies around: • End user application policies, which focus on accessing end user data, and will often call various userdom_* interfaces (which are provided through the system/ userdomain.if file). Most of these applications are inside the apps/ directory). • Administration applications, which are still user-facing, are more likely to enable interacting with system services and resources. • Services, which are generally daemonized applications, often interact mostly with their own resources and have a simpler structure. When we covered the sepolicy generate command in Chapter 14, Dealing with New Applications, we could select these types (and more) to generate a simple skeleton for those applications. Let's look into some example policies and identify useful calls that you might need when developing your own policies.

Constructing network-facing service policies Services that are network-facing (meaning they can be interacted with from outside the system itself) are the first set of services that need to be confined. Hence, building SELinux policy modules for network-facing services should be a primary focus of any Linux administrator that needs to ensure the confinement of applications that do not have a working policy yet. If we look at the OpenVPN service, then we find that there is an SELinux policy under services/openvpn.te that we can look into.

Identifying the resources that the service interacts with As a policy starts with identifying the types and other SELinux objects, we need to consider the resources on the system that the service interacts with. When comparing service policies with each other, you'll notice that the definitions are often very similar: • The main domain type and its entry point executable are declared first. Depending on the type of service, it contains a call as to how it would be started: as a system service (using init_daemon_domain()) or through the D-Bus system bus (using dbus_system_domain()).

Creating application-level policies

393

• The configuration files for the service (such as openvpn_etc_t), which could also differentiate between read-only files and read-write (such as openvpn_etc_ rw_t). • Runtime files (which are generally stored in /var/run) such as openvpn_ runtime_t. • Temporary files (which are generally stored in /tmp or /var/tmp) such as openvpn_tmp_t. • Log files (which are generally stored in /var/log) such as openvpn_var_ log_t. Each of the type declarations is followed by a call that marks the type appropriately. For instance, the logging_log_file() call will associate the type with the logfile SELinux attribute. This allows general logfile management domains to deal with the newly created resource through this attribute.

Handling the internal SELinux rules With the resources declared, we have to define the internal SELinux rules within the SELinux policy. These rules tell SELinux what the domain can do with its own resources, and how SELinux should behave when the resources are interacted with. We will generally have two sets of internal rules declared. One is the fine-grained permissions of the domain itself, such as if the domain is allowed to have any capabilities, creating sockets, and so on. The development of these rules is trial-and-error based: start with close to no permissions, see what AVC denials come up, extend the policy, and repeat. The other set of internal rules focuses on the interaction with the types declared earlier on. This not only includes which permissions the domain has (such as through the manage_files_pattern() calls) but also whether transitions have to occur. Setting the right set of transitions is one of the more important first steps to take while developing application policies because audit2allow and AVC denials generally do not consider the fact that a target resource has the wrong type assigned. So when we have a service that creates files in /tmp (which is labeled as tmp_t), we really want the target files to be labeled correctly (such as openvpn_tmp_t) and not inherit the tmp_t label from the directory: allow openvpn_t openvpn_tmp_t:file manage_file_perms; files_tmp_filetrans(openvpn_t, openvpn_tmp_t, file)

394

Using the Reference Policy

File transitions should be declared for all the resources involved. If a transition has to occur for both files and directories, you can mix the classes in a single call like so: files_tmp_filetrans(openvpn_t, openvpn_tmp_t, { file dir })

We can also tell SELinux that a transition should only occur if a specific filename is used: logging_log_filetrans(openvpn_t, openvpn_status_t, file, "openvpn-status.log")

It is really recommended to first consider the file transitions (and other resource transitions) before expanding the actual permissions for the domain to make sure that we are not tempted to allow the domain privileges to general types when that is not necessary.

Adding network related permissions While developing and expanding the policy, several core functions will be added, such as the kernel_* calls to allow processes to interact with the proc_t resources, system control settings, and more. Tools such as audit2allow will reasonably be able to deduce the right interfaces to call, although it does not hurt to review the interfaces to make sure not too many privileges are assigned. Network related permissions on the other hand might require some more attention. As we saw in Chapter 5, Controlling Network Communications, SELinux can dynamically address certain network flows based on the system configuration. It is likely that systems who do not have specific controls in place, such as labelled networking or SECMARK, will find that three interface calls could allow the application to work as intended: corenet_tcp_bind_generic_node(openvpn_t) corenet_tcp_bind_openvpn_port(openvpn_t) corenet_tcp_connect_http_port(openvpn_t)

These three interface calls allow the domain to be network-oriented (corenet_tcp_ bind_generic_node), listen to the OpenVPN port (corenet_tcp_bind_ openvpn_port), as well as connecting, as a client, to HTTP ports (corenet_tcp_ connect_http_port). But other calls exist that you might need to add, even though they are currently not detected. They might become necessary when the system is tuned further, such as adding support for labeled networking or introducing SECMARK filtering.

Creating application-level policies

395

The first set is to allow sending and receiving packets on generic nodes (hosts) and interfaces: corenet_tcp_sendrecv_generic_node(openvpn_t) corenet_tcp_sendrecv_generic_if(openvpn_t)

For NetLabel support, you might need to add support to receive labeled network packets: corenet_all_recvfrom_netlabel(openvpn_t)

For SECMARK support, you need to add support for sending and receiving SECMARK labeled packets: corenet_sendrecv_openvpn_server_packets(openvpn_t) corenet_sendrecv_http_client_packets(openvpn_t)

These calls might not show up in early tests, but could be needed later on, and it is recommended to consider the impact of labeled networking and SECMARK on your policy from the beginning.

Building the service interface methods We next focus our efforts on the interface methods. These are used to facilitate other SELinux policy modules to interact with the domain we're developing, although they can also be used to simplify policy development for your own policy. The three most common interfaces to define, and which other policy developers will assume exist, are the following: • A domain transition interface, such as openvpn_domtrans, allowing the given SELinux domain to execute the appropriate binaries or scripts and have the executed commands or applications run in our domain (and as such transition from the source domain to ours). • A run interface, such as openvpn_run, which is like the domain transition interface (and in fact will call it) but also allows our domain for the role. Without this interface, some roles might not be able to transition even if they call the domain transition interface. • An administration interface, such as openvpn_admin, which will be assigned to user roles/domains to allow them to administer our service. This will allow the user to interact with the processes of our domain (including killing the processes, tracing their actions, and so on) as well as to administer the files and resources used.

396

Using the Reference Policy

Within the interfaces, we need to declare the SELinux objects that we are going to explicitly reference. This allows the SELinux subsystem to validate whether the code is applicable or not: if the objects are not present in the current policy, then this interface is not valid and will not be used. Declaring objects is done with the gen_require() macro: interface(`openvpn_run',` gen_require(` attribute_role openvpn_roles; ') openvpn_domtrans($1) roleattribute $2 openvpn_roles; ')

Other interfaces can be added as needed. While you can add interfaces already just in case, be aware that once you define an interface it can be used by other policies, and you might not be made aware of this if you are not developing all policies yourself. If you, later on, want to change the behavior of interfaces or remove them, you might break other policies.

Addressing user applications If we develop end user applications, their structure will be very similar to those for more service-oriented applications. Content-wise, however, there are a few areas of attention to consider. Let's use the apps/thunderbird.te policy as an example: • The first thing we notice is that many resource defining interfaces are prefixed with userdom_. For instance, a temporary file is not files_tmp_file() but userdom_user_tmp_file(). This will ensure that the resources are known as user-managed temporary files and not regular system temporary files. • Another important addition is the support for the X Desktop Group (XDG) locations. The XDG locations, defined in the XDG Base Directory specification at https://specifications.freedesktop.org/basedir-spec/ basedir-spec-latest.html, standardize which end user locations are used for what purpose. For instance, inside ~/.cache, application cache data is stored, whereas configuration data is in ~/.config.

Creating application-level policies

397

The use of the XDG locations is supported through the xdg SELinux policy module, which enables support for types of user content as well: regular documents, downloads, music, pictures, and videos. This distinction allows developing SELinux policies that are only able to interact with that specific end user content and not all the user's data. For instance, the thunderbird application is able to manage download data (which by default is located in ~/Downloads) through the following: xdg_manage_downloads(thunderbird_t)

• To easily establish user content access, user applications should also call the userdom_user_content_access_template() template. This will automatically create booleans as well, which administrators can toggle. For instance, for the thunderbird SELinux policy, this will create thunderbird_manage_ generic_user_content. If set, then thunderbird can not only access the downloads-related resources but all user resources. • Another template that user application policies will need if they are graphical in nature, is xserver_user_x_domain_template(). This template will generate X server related SELinux objects for the application, and allow the application to use the graphical environment on the server. Important note The reference policy makes a distinction between regular interfaces and templates. Interfaces grant privileges to the domains and roles that are passed to it. Templates on the other hand will generate new objects, such as SELinux booleans, types, attributes, and more. Code-wise, templates cannot be part of boolean-triggered statements (as they do not just add type enforcement rules).

When the baseline of a user application policy is drafted, including the preceding templates, then expanding the policy through trial and error should suffice. Do make sure, however, that all resources on the user location for which you are testing the application are correctly labeled, as otherwise, the denials might trick you into granting more privileges to the domain than necessary.

398

Using the Reference Policy

Adding user-level policies If we want to create custom user and role policies, then the most confusing choice is the choice of user template to pick. This template creates a role and user domain with a specific purpose in mind, and grants a number of permissions by default:

Figure 15.1 – Relationship between user domain templates

The most common templates to pick for user/role policies are the following: • userdom_restricted_user_template() for (by default) unprivileged end user roles. • userdom_admin_user_template() for (by default) highly privileged end user roles. The other templates can be used as well, especially if more fine-grained controls over the roles and user domains are needed. Note, however, that the privileges assigned by the templates are mentioned as by default. If we want to create a role and user domain for administrating a specific service, then we do not want to use userdom_admin_user_ template(), as this will grant many more privileges than needed.

Getting help with supporting tools

399

As an example, consider the roles/dbadm.te SELinux policy for the database administration role. This role is based upon the userdom_base_user_template() interface to ensure minimal permissions are granted. The role is not meant to be used in a direct fashion (login), but rather transitioned toward (for instance, through the newrole command, or through well-defined role transitions within the policy).

Getting help with supporting tools There are tools out there that help in developing SELinux policies, and if needed we can build our own support tools as well. Let's see what support environments we can use.

Verifying code with selint While SELinux policies can be functionally working, validating whether the code itself is proper and follows best practices is important to ensure that the code is maintainable in the long run. One of the tools that support validating SELinux policy code is selint, as offered from https://github.com/TresysTechnology/selint. Once built and installed, selint offers insights into four main areas: • Convention checks validate whether the SELinux policy follows the reference policy convention on how code should be structured and documented. • Style checks give hints for code style that might be wrong, and where the developer might have intended a different behavior. • Warnings are triggered when the code has bad calls that might trigger runtime issues or security issues. • Errors catch construction faults that will result in compile issues or runtime issues. This allows the use of selint in automated build processes, as well as facilitating the development of policies. Calling selint is simple: $ selint minecraft.te minecraft.te: 31: (C): Permissions in av rule not ordered (signull before execmem) (C-005) minecraft.te: 118: (C): Require block used in te file (use an interface call instead) (S-001)

400

Using the Reference Policy

In this case, two convention malpractices were detected. One is in the ordering of permissions, while another has an explicit require block mentioned for a domain that is not part of that policy module.

Querying the interfaces and macros locally To help in finding the right interface or macro, we also want to quickly be able to show interface and macro information. With some shell scripting, we can create a few functions that help us along. The functions are provided as code together with this book. You might want to change the path that the POLICY_LOCATION variable points to at the beginning of the script. By default, it points to the system-installed interface and macros, but you can point it to repository checkouts as well: POLICY_LOCATION="/usr/share/selinux/devel"

Source the file to have access to the helper functions: $ source ./localfuncs

The helper functions you can use are the following: • With sefindif you can search for an SELinux interface that has a specific SELinux rule inside. You can use regular expressions to find the appropriate one. For instance, to find the interface that grants a domain the privileges to manage certificate files (for readability, we only show the interface code; it will be prefixed with the location where it found it): $ sefindif "manage.* cert_t" interface(`miscfiles_manage_all_certs',` manage_files_pattern($1, cert_type, cert_type) manage_lnk_files_pattern($1, cert_type, cert_type) interface(`miscfiles_manage_generic_cert_dirs',` manage_dirs_pattern($1, cert_t, cert_t) interface(`miscfiles_manage_generic_cert_files',` manage_files_pattern($1, cert_t, cert_t) manage_lnk_files_pattern($1, cert_t, cert_t)

Getting help with supporting tools

401

• With seshowif the interface in its entirety (excluding the comment) is displayed. For instance, to show the miscfiles_manage_all_certs() interface, use the following code: $ seshowif miscfiles_manage_all_certs interface(`miscfiles_manage_all_certs',` gen_require(` attribute cert_type; ') allow $1 cert_type:dir list_dir_perms; manage_files_pattern($1, cert_type, cert_type) manage_lnk_files_pattern($1, cert_type, cert_type) ')

• With sefinddef and seshowdef, the same is possible but for the supporting macros. For instance, to see the content of the admin_pattern() helper macro, use the following code: $ seshowdef admin_pattern define(`admin_pattern',` manage_dirs_pattern($1,$2,$2) manage_files_pattern($1,$2,$2) manage_lnk_files_pattern($1,$2,$2) manage_fifo_files_pattern($1,$2,$2) manage_sock_files_pattern($1,$2,$2) relabel_dirs_pattern($1,$2,$2) relabel_files_pattern($1,$2,$2) relabel_lnk_files_pattern($1,$2,$2) relabel_fifo_files_pattern($1,$2,$2) relabel_sock_files_pattern($1,$2,$2) ')

While such functions do not offer the same versatility as a full-fledged policy editor suite would, they can help in quickly finding the right interface or macro.

402

Using the Reference Policy

Summary The reference policy is the most common source for SELinux policy development, and with years of development effort and maintenance, it has grown to be a full policy set with a vivid development community, and active support by various tools (including audit2allow, as well as the selint application). We've learned how policies are generally structured, and how to start building SELinux policy modules for the most common use cases: application services, end user applications, and user roles. To help us in developing these policies, we've seen that selint can do code-style analysis, whereas some shell scripts can help us parse the interface files for quick help. In our final chapter, we will look into CIL style SELinux development.

Questions 1. Why don't Linux distributions use the reference policy natively? 2. What are the three main policy files needed for an SELinux module, and what is their purpose? 3. Why is a permission set such as exec_file_perms preferred over explicitly listing the permissions? 4. What is the difference between interfaces and templates? 5. Why is the database administration role defined in dbadm.te not using userdom_admin_user_template?

16

Developing Policies with SELinux CIL While the reference policy is the most frequently used language and development style for SELinux policies, the Common Intermediate Language (CIL) is a powerful, but more low-level language construct to use to develop SELinux policies. Low-level as it might be though, it is still very much readable and well supported, as SELinux tools will use CIL under the hood when using other languages. Since CIL is the main language used, we know it can be used to build entire policies. Sadly, there are no supporting constructs available for developers to use, unlike the reference policy. However, we can still learn how to customize the current policy, creating specific definitions that are not possible with the more common reference policy, and even build a complete application policy if we choose. In this chapter, we're going to cover the following main topics: • Introducing CIL • Creating fine-grained definitions • Building complete application policies

404

Developing Policies with SELinux CIL

Technical requirements Check out the following video to see the Code in Action: https://bit.ly/3dLYP2Q

Introducing CIL CIL has been designed to be the main language to have policies built in, and is the lowest readable format. After CIL, the SELinux code is transformed in binary to send off to the Linux kernel (and SELinux subsystem) for loading in memory. Administrators might be inclined to think that the binary files, generated when building a SELinux policy module using the reference policy method, are the final binaries. However, as we've seen in Chapter 1, Fundamental SELinux Concepts, the semodule command converts and translates this into CIL before building the final format. Let's see how these translations work and what we can learn from them.

Translating .pp files to CIL When a non-CIL SELinux policy module is loaded, the semodule command is designed to first consider the module as an unknown format, and extract the High Level Language (HLL) information from it. HLL is an abstract term that the SELinux utilities use to define any SELinux source format that it knows how to convert to CIL later on. Currently, only .pp files are supported as an HLL. As the HLL format of the SELinux policy modules built by reference policy (or the other classic SELinux development) is the same as the module generated, this phase often just involves creating a copy. We can see this when we compare an HLL file with the original: $ cp /var/lib/selinux/targeted/active/modules/400/pgpool/hll pgpool.hll.bz2 $ bunzip2 pgpool.hll.bz2 $ sha512sum pgpool.pp pgpool.hll b81ba4ac...c0db pgpool.pp b81ba4ac...c0db pgpool.hll

Once it has converted or extracted the data, semodule will convert it to CIL code. For each supported HLL, a convertor is available in /usr/libexec/selinux/hll. Currently, only the pp command is available, which is used to convert this older style into CIL.

Introducing CIL

405

Let's see this in action: $ /usr/libexec/selinux/hll/pp pgpool.pp (type pgpool_t) (roletype object_r pgpool_t) (type pgpool_exec_t) ... (filecon "/var/run/pgpool(/.*)?" any (system_u object_r pgpool_ var_run_t ((s0) (s0))))

So, in essence, when we are developing reference policy style modules, they will be converted into CIL anyway. The resulting CIL code, however, does not have any facilitating constructions inside. For instance, all permissions are expanded, and all interactions with other resources or types outside of the SELinux policy module are listed as well. There are no longer any supporting macros or interfaces. In the Building complete application policies section, we'll see that CIL does support abstractions, so the current observation is only due to the translation that the pp command performs.

Understanding CIL syntax When we develop CIL, the most obvious observation is that it likes brackets. CIL uses S-expression syntax, popularized by Lisp, which results in tree-structured data. The first identifier in an expression that tells CIL what the construction is about. Let's take a look at the last statement we received when converting the pgpool.pp binary into CIL, now formatted for convenience: (filecon "/var/run/pgpool(/.*)?" any ( system_u object_r pgpool_var_run_t ( (s0) (s0) ) ) )

406

Developing Policies with SELinux CIL

If we look at this statement in detail, we can deduce the following: • We have a filecon statement, which takes three arguments: the path expression, the type of resources to which it applies, and the SELinux context to associate with it. • The SELinux context has four fields associated with it: the SELinux user, the SELinux role, the SELinux type, and the SELinux sensitivity range. • The SELinux sensitivity range has two values, a low-end and a high-end value. This statement is equivalent to what the reference policy would define in the file context part of the policy module (the file with the .fc suffix) like so: /var/run/pgpool(/.*)? var_run_t,s0)

gen_context(system_u:object_r:pgpool_

Luckily, we do not need to seek and interpret the code just to understand and see what is going on. The SELinux project has extensive CIL documentation available, explaining how the language works and what it all supports. The information is available at https:// github.com/SELinuxProject/selinux/tree/master/secilc/docs. Keep in mind though that CIL policy development is still in its infancy, so coverage of the CIL constructs that are not used by the HLL conversion mechanics is very low. Let's now see what we can do with CIL.

Creating fine-grained definitions Throughout this book, most small SELinux policy adjustments have been made using CIL. These are small, fine-grained definitions that require little development effort, and have the benefit of being directly loadable.

Depending on roles or types The CIL language requires some order in how types or roles are linked in the policy. Sometimes, when we develop CIL policies, the order of the types might not be addressed properly. To work around this issue, a default attribute called cil_gen_require is used. When types or roles are assigned to the cil_gen_require attribute, they are automatically linked correctly in the policy. This is not a CIL requirement though, but a convention that the SELinux utilities use. The attribute actually exists twice, once as a type attribute and once as a role attribute.

Creating fine-grained definitions

407

They might have the same name, but are two different attributes: (roleattributeset cil_gen_require system_r) (typeattributeset cil_gen_require direct_run_init)

The functions used, roleattributeset and typeattributeset, assign the second argument (which is the attribute name) to the third argument (which is the role or type). Roles or types can be attributes themselves, as shown for the direct_run_init attribute. For developers, it is recommended to always have the types or roles that you are going to use made part of the cil_gen_require attribute, and to use attributes as much as possible to simplify development activities. Rather than granting all possible allow rules to each and every domain that can interact with your policy resources, grant them to an attribute, and then assign this attribute to the domains. This creates a much smaller policy and is easier to maintain.

Defining a new port type When more intricate SELinux policies are developed, there are a few settings that we cannot add in a SELinux module—at least not when using the traditional or reference policy style coding. When we want to use these, we need to rebuild the entire policy and make the adjustments in the so-called base policy; the main and first policy loaded before the modules are added. This, however, requires access to the full SELinux policy sources and a process to use them (as you will overwrite the Linux distribution's SELinux policy and should make sure that any system update does not overwrite your policy again). One way to establish whether a statement is supported in a SELinux module, besides just testing it out, is to look at the online documentation. As an example, let's take the port declaration statement portcon, which is part of the network-oriented statements. This statement is documented on https://selinuxproject.org/page/ NetworkStatements, where we can see that portcon is not valid in a module policy, nor can it be toggled through a SELinux Boolean. Luckily, this is not the case when using CIL. Let's create a custom port type, for example, pgpool_port_t, and map it to a free port, say, TCP port 50123: ; Port type (type pgpool_port_t) ; Some attributes to match the current SELinux policy requirements (typeattributeset defined_port_type pgpool_port_t) (typeattributeset reserved_port_type pgpool_port_t)

408

Developing Policies with SELinux CIL

(typeattributeset port_type pgpool_port_t) ; Our dependency mappings (roleattributeset cil_gen_require object_r) (typeattributeset cil_gen_require pgpool_port_t) ; Make sure object_r is allowed for pgpool_port_t (roletype object_r pgpool_port_t) ; The port mapping itself (portcon tcp 50123 (system_u object_r pgpool_port_t ((s0) (s0))))

Before we load the policy, we can clearly see that this port is not an assigned one: # seinfo --port 50123 | grep tcp portcon tcp 32768-60999 system_u:object_r:ephemeral_port_t:s0

As we've seen throughout this book, we can load this policy file immediately, without having to build or compile it: # semodule -i pgpool_port.cil

After loading, the type is assigned to the port, and we can use it to fine-tune our SELinux policies: # seinfo --port 50123 | grep tcp portcon tcp 50123 system_u:object_r:pgpool_port_t:s0 portcon tcp 32768-60999 system_u:object_r:ephemeral_port_t:s0 # semanage port -l | grep 50123 pgpool_port_t tcp 50123

Many of the constraints that policy developers had when using the traditional SELinux style development no longer apply to CIL. As the SELinux utilities convert all high-level constructs to CIL, it is possible that SELinux developers might remove these constraints altogether, although this has to be carefully assessed to make sure no unwanted side effects arise.

Adding constraints to the policy Another area that was not accessible for regular policy developers was to add constraints to the policy. Constraints limit actions based on the entire SELinux context (and not just types), and they are the closest thing we can find to negating existing rules.

Creating fine-grained definitions

409

Important note We do not recommend adding constraints to an existing policy just to work around rules that we don't like. Constraints are not visible in regular policy queries, as with sesearch. Administrators might be very confused when sesearch indicates that an action is allowed while the system is refusing to allow it.

With CIL, we can add constraints to a live policy to do exactly that. Keep in mind though that constraints don't actually allow anything – they merely put limits on what the SELinux subsystem will see as a valid action. A constraint statement that supports reading any possible type does not actually allow this, as there still need to be type enforcement rules in place to actually allow these actions. For instance, if we want to remove the staff_t domain's ability to read the /etc/ passwd file (which has the SELinux type, passwd_file_t), then we can add in a constraint that supports the reading of all possible types, unless the source domain is staff_t, in which case we support the reading of all possible types except passwd_ file_t: (constrain (file (read)) (or (and (eq t1 staff_t) (not (eq t2 passwd_file_t)) ) (not (eq t1 staff_t)) ) )

Once loaded, we can confirm that the constraint is active: # seinfo --constrain | grep passwd_file_t constrain file read (t1 == staff_t and not ( ( t2 == passwd_ file_t ) ) or not ( t1 == staff_t ) ));

And indeed, trying to read the passwd file is prohibited: $ cat /etc/passwd cat: /etc/passwd: Permission denied

410

Developing Policies with SELinux CIL

To show how the use of constraints is confusing for administrators, let's see what sesearch has to say on this: # sesearch -A -s staff_t -t passwd_file_t -c file -p read; allow nsswitch_domain passwd_file_t:file { ... read }; allow staff_t passwd_file_t:file { ... read };

So, while the policy has two rules that would allow it (one for the nsswitch_domain attribute, and one explicitly for the staff_t domain), the constraint has limited this action, but this is not obvious from the sesearch output.

Building complete application policies We can build complete application policies with CIL as well. However, keep in mind that there are no interfaces or support macros out there that we can use to rapidly develop policies. Furthermore, there are no templates or suchlike available to jumpstart such initiatives. But that shouldn't stop us, and it will allow us to show a few more details of the CIL language. We will also see that the CIL language does support interface constructs (they are even recommended), but the community has not yet fully embraced it through a reference policy-like project.

Using namespaces The CIL language supports namespaces, which allows for a higher flexibility in developing policies. The generated CIL policies always use the main, global namespace, so we will not find examples of namespaces in the generated policies. We can, however, show how this works easily. Let's create a skeleton file that will contain our CIL-developed pgpool policy: ; Dependencies ; Pgpool support (block pgpool ; Declarations (type domain) ; Local policy ; Behavior

Building complete application policies

411

; File contexts )

The namespace created in the preceding code is the pgpool namespace, identified through the block statement. Namespaces in CIL are hierarchical. If we want, we could create a namespace within pgpool by nesting another block statement within. When we encounter namespaces in general constructions, we need to use the dot separator. In the example, we have defined a type called domain inside the pgpool namespace. If we would want to query it later through sesearch, the full name would be pgpool.domain (which would have the same purpose as the pgpool_t name in a more classically developed policy). Informational note In subsequent code listings of the policy, we will only show the added and most relevant statements rather than including all previously added statements. Without supporting interfaces (which CIL calls macros), the policy file will quickly become quite large, which does not aid readability. By focusing on the relevant and added statements, the development pattern for CIL policies is easier to explain.

As the main policy has all objects within the global namespace, we will need to refer explicitly to this global namespace. This is done by prefixing the name with a dot. For instance, if we want to assign the pgpool.domain type to the system_r role, our policy needs to be adjusted with the following: (roleattributeset cil_gen_require system_r) (block pgpool (type domain) (roletype .system_r domain) )

Here, the roletype statement is used to assign the domain type (defined in the pgpool namespace) to the system_r role (defined in the global namespace).

Extending the policy with attribute assignments When we develop a policy, it is recommended to use attributes as much as possible. Many attributes will automatically grant the necessary privileges to jumpstart a policy development, reducing the number of allow statements that need to take part of the policy.

412

Developing Policies with SELinux CIL

The main reference policy, which is still in use on the system as a whole, defines quite a few attributes, as seen in all previous chapters. So, let's assign the daemon attribute to our domain: (typeattributeset cil_gen_require daemon) (block pgpool (type domain) (typeattributeset .daemon domain) )

By assigning the daemon attribute, all existing policy rules for daemons are automatically applied to the pgpool.domain SELinux domain. To find out which attributes are sensible to add, we can take a peek at existing daemon domains: # seinfo -t postgresql_t -x Types: 1 type postgresql_t, nsswitch_domain, can_change_object_ identity, corenet_unlabeled_type, domain, kernel_system_state_ reader, netlabel_peer_type, daemon, syslog_client_type, pcmcia_ typeattr_1;

Now, we do not need to blindly take up all attributes, starting instead with those we feel confident with.

Adding entry point information Our next step is to add entry point information to the policy. This is a necessary step before we can start testing out, because we want the domain to become active. For that to happen, it has to be executable by the init system (or systemd) and transition to the domain we've just declared. Let's start by defining our entrypoint type (pgpool.exec) and associate it with the right attributes: (roleattributeset cil_gen_require object_r) (typeattributeset cil_gen_require file_type) (typeattributeset cil_gen_require direct_init_entry) (block pgpool (type exec) (roletype .object_r exec) (typeattributeset .file_type exec) (typeattributeset .direct_init_entry exec)

Building complete application policies

413

(allow domain exec (file (entrypoint ioctl read getattr lock map execute open))) (typetransition .initrc_domain exec process domain) )

In this code block, we've performed several steps to ensure that a transition will occur: • We've associated the pgpool.exec type with the file_type attribute (which is a generic attribute for files) and the direct_init_entry attribute (which is for file types that are used to launch system services). • We've marked the pgpool.exec type as entrypoint for the pgpool.domain type, as well as granted this domain the necessary privileges to read, open, and execute the pgpool.exec labeled resources (as needed for a starting process). • We've declared a type transition so that any initrc_domain labeled process that executes the pgpool.exec labeled resource will result in a domain transition toward pgpool.domain. We can now finish this step by adding a file context definition: (block pgpool (filecon "/usr/bin/pgpool" file (.system_u .object_r exec ((s0) (s0))) ) )

With these changes made, we can load the policy and relabel the file: # restorecon -v /usr/bin/pgpool Relabeled /usr/bin/pgpool from system_u:object_r:bin_t:s0 to system_u:object_r:pgpool.exec:s0

We can now attempt to start the pgpool service, and hope that it fails (as that will show that the transition was successful, given that the pgpool.domain SELinux domain hardly has sufficient privileges to successfully start the entire service).

Gradually extending the policy further Once the domain transitions are successful, we can gradually extend the policy further through trial and error, just like we would do when developing SELinux policies using the reference policy style. However, rather than using audit2allow to guide us, we will need to interpret the denials ourselves and see how to better approach it.

414

Developing Policies with SELinux CIL

Consider the failures that appear after starting the service: # ausearch -i -m avc -ts recent (Output reformatted for readability) avc: denied { map } for scontext=pgpool.domain tcontext=ld_so_t tclass=file avc: denied { read write open } for scontext=pgpool.domain tcontext=null_device_t tclass=chr_file

Now, rather than immediately adding allow rules for these types, let's see how this is accomplished for other daemons on the system: # sesearch -A -s postgresql_t -t ld_so_t -c file -p map allow domain file_type:file map; [ domain_can_mmap_files ]:True allow domain ld_so_t:file { execute getattr ioctl map open read };

So, this permission is based on the domain attribute, which we indeed forgot to add to the policy. Let's rectify this and retry: (typeattributeset cil_gen_require domain) (block pgpool (type domain) (typeattributeset .domain domain) )

In this example, we can also see clearly what the impact is of the namespaces within CIL. We assigned the (global namespace-hosted) domain attribute to the (pgpool namespace-hosted) domain type. They are both named domain, but have a different namespace. This also shows how important attributes are. Of course, not all privileges can be granted through attributes. By adding the target types as a dependency, we can directly include allow statements in our policy, like we did with the entrypoint declaration. For instance, if we would want to explicitly allow our domain to signal the postgresql_t domain, execute the following command: (typeattributeset cil_gen_require postgresql_t) (block pgpool (allow domain .postgresql_t (process (signal))) )

Building complete application policies

415

As we are adding more and more privileges to the policy, we might want to optimize some of the definitions. There are two optimizations supported by CIL, and they are, not unsurprisingly, aligned with the reference policy's simplifications as CIL was developed by the same community.

Introducing permission sets The first simplification we can do is to simplify the permission sets we use. Remember the allow rule we added to allow our domain to execute its entrypoint file: (allow domain exec (file (entrypoint ioctl read getattr lock map execute open) ) )

Were we to use a reference policy-style approach, we would combine many of these permissions through the exec_file_perms macro. Well, CIL supports something similar, through a statement called classpermissionset. If we want to simulate the reference policy-style approach completely, we would define classpermissionset in the global namespace, and use it, as follows: (classpermission exec_file_perms) (classpermissionset exec_file_perms (file (ioctl read getattr lock map execute open))) (block pgpool (allow domain exec (file (entrypoint))) (allow domain exec exec_file_perms) )

In this example, we've defined classpermissionset in the global namespace, and then referred to it. Unlike the reference policy, however, we cannot just add exec_file_ perms inside the permissions together with entrypoint. The classpermissionset statement has an explicit reference to the class associated with it. The allow statement in CIL is therefore a separate one that does not contain a class reference itself. Furthermore, in the example, you will also notice that we did not prefix the exec_file_ perms name with a dot, to refer to the global namespace. While we can prefix it perfectly to be consistent with the rest of the policy, using a dot prefix is not mandatory if there is no possible collision. If no local definition for a name exists within the current namespace, the policy will then check whether the parent namespace (and, hence, also the global namespace) has the name defined.

416

Developing Policies with SELinux CIL

So, while the preceding policy will work just fine, we do recommend to prefix the global namespace-oriented names with a dot to make sure no local override would confuse the policy later on.

Adding macros The final simplification we can introduce is to add macros. CIL has an explicit support for macros, which allows them to be part of the loaded policy, and not just be referred to on the filesystem. With CIL macros, the code is part of the policy itself. There is no need to refer to the CIL code while building policies. While this is a best practice that is aligned with object-oriented programming (as we can add macros to our namespaces so that they remain within the same object), the downside is that the current SELinux utilities are not able to quickly show which macros (and which interface they require) are available in the policy. Now, let's enhance our pgpool policy with a domain transition macro, similar to the pgpool_domtrans() interface that would be created through a reference policy-style development: (block pgpool (macro domtrans ((type SOURCEDOMAIN)) (allow SOURCEDOMAIN exec exec_file_perms) (allow SOURCEDOMAIN domain (process (transition))) (typetransition SOURCEDOMAIN exec process domain) ) )

The macro definition itself starts with a name (in our case, domtrans) followed by the interface. This interface defines how many arguments are passed to the macro, and which type they have. In our example, only one argument is passed, and it is a SELinux type. The macro is then followed by the code that is applied. The argument itself is referenced in the code (SOURCEDOMAIN) and will be substituted with the argument that is given later on, when the macro is explicitly called. While our example uses a capitalized variable name, this is not mandatory, and only serves as a visual statement of what will be substituted.

Building complete application policies

417

In another CIL policy, we can refer to this macro through the call statement. For instance, to allow the postgresql_t domain to transition to the pgpool.domain SELinux domain, we would add the following call statement to our policy: ; Equivalent to "pgpool_domtrans(postgresql_t)" in refpolicy (typeattributeset cil_gen_require postgresql_t) (call pgpool.domtrans (postgresql_t))

CIL macros provide all that is needed to generate the same simplicity in developing SELinux policies as we have within the reference policy, and even more as there are many constraints not applicable to CIL policies. While it is possible that this will happen in the future, it is not planned at this moment for a number of reasons: • The current reference policy has a significant amount of code in it, which would all need to be reworked. Furthermore, Linux distributions have extended this policy with many of their own additions, so the work needed to rewrite the SELinux policy code into CIL is significant. Not impossible, but not a feat to accomplish in a few weeks. • Almost all of the information and documentation online that helps developers in writing SELinux policies is based upon the current reference policy. This major source of information would become stale the moment a switch occurs, and the amount of documentation available online for CIL-based policy development is still pretty slim. • The CIL policy, while very powerful, is also a bit more complex due to its S-expressions. The design intention of CIL was not to replace SELinux policy development with CIL, but to allow higher level languages to be developed that translate and convert into CIL easily. Hence, if a rework is going to be done anyway, it is much more likely that a user- and development-friendly language will be designed that can be easily converted into CIL. As the SELinux development progresses, both on the policy level as well as in terms of user space and kernel support, we can expect more additions to be added to CIL and to its supporting tools.

418

Developing Policies with SELinux CIL

Summary CIL for SELinux is a powerful, lower-level syntax and language that is used to express all possible SELinux policy code. The SELinux userspace utilities will automatically convert existing policies into CIL code, but through this conversion, a lot of CIL constructs are not used: the conversion only uses a smaller set of CIL capabilities to establish a valid translation. The more advanced CIL capabilities, such as namespace support, macros, and the permission sets through the classpermissionset statement, are useful when developing our own, CIL-based SELinux policies. In this chapter, we've learned how to use CIL to build complete application policies. Because there is no reference policy-like framework to simplify development, we had to write all of the necessary code constructs ourselves. While this means that developing CIL-based policies is more resource intensive, we did also see that CIL has a few benefits that reference policy-style development cannot deal with, such as the ability to declare ports or add SELinux constraints to an active policy. We ended the chapter with a brief overview of why CIL-based development is not more widely used, but we will notice continuous improvements within SELinux on this matter in the foreseeable future. This concludes our book and the information we have to offer to you. However, it is only the start of a journey, not the end. SELinux is a widely used technology, and we hope that this book provides you with the right material and knowledge to understand, grow, and contribute to the ecosystem. Thank you for your interest and your dedication.

Questions 1. How do we know CIL is here to stay? 2. Is the cil_gen_require attribute mandatory for CIL development? 3. What are examples of declarations that developers can do with CIL but not with other SELinux language styles? 4. How can we create similar support constructions such as interfaces in CIL?

Assessments Chapter 1 1. The most important difference is that, with a DAC system, the user has full control over who gets which kind of access to the user's data. It is left to the discretion of the user, hence the name. With MAC systems, the system administrator (or security administrator) defines how accesses are handled and enforced. Access is mandated by a policy, and users cannot work around this if the administrator does not allow it. 2. Linux has introduced hooks inside its kernel code, which developers can subscribe to with their own code. These hooks are part of the Linux Security Module (LSM) framework, an extensible framework that is natively part of the Linux kernel. SELinux is one of the MAC technologies that use this LSM framework (and the hooks it offers) to provide mandatory access control capabilities to the Linux kernel and its applications. Other technologies also exist, including AppArmor. The SELinux subsystem code itself is also made part of the main Linux kernel, as are the main other LSM implementations, although this is not a mandatory requirement for LSM-capable technologies. It does, however, support the notion that SELinux is a well-established, open source technology. 3. The four fields of an SELinux context are as follows: the SELinux user, the SELinux role, the SELinux type, and the sensitivity level (or the sensitivity range). The sensitivity level might not always be present: Linux distributions might opt to disable support for sensitivities in their policies. In that case, the SELinux context will only have the first three fields. 4. SELinux has the concept of a role, which SELinux types can be associated with. As SELinux mostly focuses on the types to handle its enforcement (SELinux is mostly a type enforcement system), the role-based access control is implemented by limiting the types that a role can be associated with.

420

Assessments

A user that has a DBA-related role will only be able to interact with the system from within DBA-associated types. As that role does not have any associations with other types, the user cannot acquire the privileges of these other types either. 5. While there is a project called the reference policy, most Linux distributions will deviate from this policy for several reasons. The main reason why no single SELinux policy exists is because SELinux is a fine-grained system, and thus can be tweaked and adjusted to fit the design and usage principles of a Linux distribution. Asking why no single SELinux policy exists for all Linux distributions is almost the same as asking why there are multiple Linux distributions. Each distribution has its own focus, design, principles, and decisions behind it, and the SELinux policy needs to be aligned with these in order to be successful.

Chapter 2 1. Administrators should first analyze the situation to see why a problem is being triggered. Perhaps the problem is due to an incorrectly assigned context, or the process has not been started using the correct methods. If the denial itself were to be allowed, administrators should create an update to the SELinux policy (just like they would update firewall rules as required). If this is not feasible, then administrators should consider putting SELinux in permissive mode, but only for that particular application that is causing problems. If that is also not feasible, then administrators should put the system in permissive mode, but making sure that this is accepted by the organization and security principles of the environment. Only if even this is not feasible or solves the problem should an administrator shake their head, curse the higher powers, and disable SELinux. 2. If the system has the audit daemon running, then SELinux logging will be part of the audit logs. They can be displayed using tools such as ausearch, or read directly from the system at /var/log/audit. If no audit daemon is running, then the SELinux log events will be picked up by the system logger or be available through the kernel ring buffer. The kernel ring buffer can be read using the dmesg command. If the system logger is picking up the events, they will most likely reside in /var/log/messages.

Chapter 2

421

3. Applications that actively query the SELinux policy or SELinux system will link with the libselinux library. If that is the case, then this can be seen using readelf, ldd, or objdump, showing that /lib64/libselinux.so.1 (or similar) is used: $ ldd /bin/ls | grep selinux libselinux.so.1 => /lib64/libselinux.so.1 (0x00005d415f3f03f0)

While applications could build statically (meaning they include the necessary code in their final build and will not show any dynamic linking with the libselinux library), this is more the exception than the rule for most Linux systems. 4. The AVC, or Access Vector Cache, is a cache that contains the recent and most frequently used enforcement checks, allowing the SELinux subsystem to query more rapidly whether an action can be granted. Without the AVC, the SELinux subsystem would need to go through the entire policy over and over again for each action that is taken on the system. Suffice to say that this would slow down the system tremendously. 5. No, there are a couple of other log events that administrators should look out for when specifically dealing with SELinux. One is USER_AVC, which is used for AVC-like events, but triggered from an application that uses the SELinux policy, and does the enforcements itself (rather than through the Linux kernel). The other one is SELINUX_ERR, which is used when an internal error or violation is triggered that is not related to regular type enforcement. Other event types that are closely related, but are not exclusive to SELinux, exist as well. For instance, MAC_POLICY_LOAD, MAC_POLICY_CHANGE, and MAC_ STATUS are events that are triggered whenever a MAC system state or policy is changed.

422

Assessments

Chapter 3 1. There is an intermediate step needed to associate a role with a Linux account, and that is the SELinux user. A Linux account (or login) is mapped to an SELinux user. The SELinux user is then mapped to one or more SELinux roles that that SELinux user can be in. If we want to assign an additional role to a Linux user, we need to add it to the SELinux role that that Linux account is mapped to. However, if more Linux accounts are mapped to the same SELinux user, then we first need to make sure that all these accounts are indeed allowed to use this role. If not, a dedicated SELinux user has to be created for the Linux account. 2. Yes, the mappings are considered when a user logs in through a particular service. It is possible for administrators to tune the mappings to be dependent on the service, as seen in the Customizing logins toward services section. 3. Most SELinux domains do not allow the SELinux user of a context to be changed. This allows the tracking of activities based on the SELinux user, even when the regular Linux user has changed their user ID. Note that this is not exclusive to SELinux, however. Linux does support a distinction between the real user ID (which stays static as much as possible) and the effective user ID (which can change, for instance, when executing a setuid application). SELinux users also allow granularity as regards the SELinux policy, for instance, when using user-based access control. In that case, SELinux users cannot access resources that are owned by a different SELinux user. 4. PAM is a flexible, modular system that Linux uses to authenticate users. Rather than having all the different technologies and services on a system implement authentication over and over again, they use PAM to handle the authentication flow. Administrators only then need to focus on PAM or PAM-related configurations to ensure that their systems are properly accessed. For SELinux, PAM is needed to allow the authentication to check the mappings (between Linux users and SELinux users), which is supported through pam_ selinux.so.

Chapter 4

423

Chapter 4 1. The most common option is -Z, and is supported by tools such as ls, mv, and ps. The same character is also used by systemd's tmpfiles application to explicitly set SELinux contexts on resources. However, while this is the most commonly used option, not all tools follow this convention, so we recommend to always consult the tool's help or manual page. 2. In most cases, the context is stored as an extended attribute of the file or directory within the filesystem. This extended attribute is the security.selinux attribute, and can be queried with tools such as getfattr or stat. However, not all filesystems support extended attributes. In that case, the SELinux context is obtained through the mount options of that filesystem, and all resources on the filesystem then use the same context. 3. The chcon application directly alters the SELinux context for a file, but does not adjust the system's file context definitions. If, at any point in time, the system or an administrator relabels the file or the entire filesystem (which is a common remediation for SELinux issues), then the SELinux context of the file is changed back. Hence, chcon is only recommended for transient SELinux context changes or to validate whether a context change solves an issue. Once confident that the new context is needed, it should be registered in the system's file context definition through semanage fcontext. 4. Yes. While SELinux's tools have the concept of most specific rule wins for the context definitions provided by the Linux distribution, this concept does not apply to file context definitions that are local to the system (in other words, executed by the system administrator). For locally defined file context definitions, the first rule that matches a patch will be used, regardless of the context definitions that follow. 5. If you only want to relabel a selected set of files, such as recursive applications to a given directory, the restorecon command should be used. If the entire filesystem needs to be relabeled, either use fixfiles (CentOS and related distributions) or rlpkg (Gentoo). Another method is to create an empty file called /.autorelabel and reboot the system. The system will detect this file, relabel the entire filesystem, remove this file, and then reboot again.

424

Assessments

6. The source domain needs the transition privilege vis-à-vis the target domain. It also requires the execute privilege on the executable file. This executable file has to be marked as an entry point for the target domain. Finally, the role for which a transition is to occur has to have the target domain as an allowed type. 7. Multiple SELinux types can be assigned an SELinux attribute, and the SELinux policy can then use this attribute as a source or target for its rules. Such attribute-based rules then automatically apply to all types assigned from this attribute.

Chapter 5 1. The command to apply a type to a TCP port is created with semanage. For instance, to apply the ssh_port_t type to TCP port 10122, execute the following command: # semanage port -a -t ssh_port_t -p tcp 10122

However, this only works as long as the port itself is not already explicitly mapped to an SELinux type. You can query whether this is the case with sepolicy, for example: # sepolicy network -p 10122

If the port is part of an unreserved range, then it can be altered. 2. No, SECMARK is local to the system. Once a network packet is received by the Linux host, the SECMARK rules will associate a label with that network packet, but this label is only retained in memory on the system itself. Once a packet leaves the Linux system, it will not show any trace of SECMARK labeling. 3. The subcommands used by semanage are ibendport (to apply a label or sensitivity to an InfiniBand network port) and ibpkey (to apply a label or sensitivity to a partition key). 4. While labeled IPsec itself does not require specialized equipment, it does require all participating hosts to have the same view on what each label implies. This means that all hosts, in the case of SELinux-based labeling, need to have SELinux active, and preferably with exactly the same SELinux policy.

Chapter 6

425

Chapter 6 1. Ansible (using setype within the file module) and Puppet (using seltype in its file module) are the only two tools that have native support for explicitly setting SELinux contexts on resources. However, Chef will automatically relabel resources according to the defined file context rules, but you cannot natively override this behavior. 2. Except for SaltStack, all orchestration tools have support for community-built and community-supported modules that extend native support of the tools. Ansible's Galaxy, Puppet's Forge, and Chef 's Supermarket are the main communities for these customizations. All orchestration tools (including SaltStack) are flexible enough to use commands and simple checks to check state and make changes, effectively allowing administrators to customize the definitions to their liking. 3. All tools have their own view and design on how they approach things. Ansible, for instance, pushes its changes to the remote nodes, whereas the others generally use remote agents to connect to the central system to obtain the latest changes. SaltStack even supports both approaches. While all tools have some SELinux support included, some have many more SELinux features enabled out of the box than others. Luckily, through the use of the community-supported additions (modules), the SELinux support for almost all of the orchestration tools can easily be augmented.

Chapter 7 1. The unit files in /usr/lib/systemd/system are managed by the Linux distribution itself. Whenever a new update to the software is deployed on the system, these files are overwritten. Modifications to unit files should be placed in /etc/systemd/system instead, as they overrule the settings in /usr/lib/systemd, and software deployments should not place any of their unit files in that location. 2. The application is tmpfiles, and is part of the systemd suite. To have it reset a context, a configuration file has to be created (in /etc/tmpfiles.d for locally defined changes) and use the z directive (to reset the context of a single file) or the Z directive (to recursively set the context of an entire directory).

426

Assessments

3. The journalctl command allows filtering on variables that it obtained from the event itself. One of these variables is the SELinux context of the service that generated the event. To filter on a particular value, you use the variable name as an argument to the journalctl command like so: # journalctl _SELINUX_CONTEXT=system_u:system_r:init_t:s0

If you do not know what SELinux context to use, then the Bash completion might be of assistance. Just declare _SELINUX_CONTEXT= in the preceding command, and then press Tab twice to see all the valid values. 4. If the SELinux policy itself does not have a proper named file transition rule in place (which would automatically have the node created with the correct SELinux context), then you can tell udev to do this for you. Find the udev rule that would create the device node for the device, and copy this rule into /etc/udev/rules.d. Then, update the rule by adding a SECLABEL{selinux}= action, like so: KERNEL=="fd0", ..., SECLABEL{selinux}="system_u:object_r:my_device_t:s0"

Such rules have to be placed in /etc/udev/rules.d rather than /usr/lib/ udev/rules.d as the latter location is managed by the distribution, and new installations or updates will overwrite the files located therein. 5. No. The SELinux policy is only checked by D-Bus if the D-Bus policy file itself refers to an SELinux context (using the busconfig > selinux > associate XML entities). If no SELinux mapping is defined in the policy, then D-Bus cannot know which association to validate. This is unlike the message flows, however, which are immediately governed by D-Bus through the SELinux policy. 6. Apache can be made SELinux-aware because it has a modular design, and allows third-party modules to be applied to its own environment. While no SELinux support is enabled within the core Apache code, additional modules (such as mod_ selinux) can be added that do enable SELinux support.

Chapter 8

427

Chapter 8 1. It is, although it is not natively enabled. SEPostgreSQL is offered through one of the additionally supplied modules within PostgreSQL called sepgsql. As such, it is part of the default technology, but not enabled by default. 2. As the sepgsql module requires a session context, the PostgreSQL database needs to either only be accessed from the local system (using the Unix domain sockets), or labeled networking needs to be enabled and set up in the network. Without labeled networking, any remote connection to the database will fail to provide any context information, and sepgsql will refuse the connection. 3. When a database object is created in PostgreSQL, it will automatically receive an SELinux label. Administrators or database owners can change the labels using the SECURITY LABEL statement in PostgreSQL: db_test=# SECURITY LABEL ON COLUMN tb_users.phash IS 'system_u:object_r:sepgsql_secret_table_t:s0';

To query the current label, consult the pg_seclabels table in PostgreSQL: db_test=# SELECT objname,provider,label FROM pg_seclabels WHERE objname='tb_users.phash';

If you do not know the object name by heart, use LIKE and use % as a glob character. 4. The sepgsql module does not interact with the Linux audit subsystem, relying instead on the logging capabilities and interface used by PostgreSQL. As a result, any decision logging that sepgsql does will be found in the PostgreSQL system logs.

Chapter 9 1. The unique idea that sVirt has that differentiates it from a more standard SELinux configuration is to use SELinux's MCS support to the next level. By randomly assigning two categories to a guest, sVirt can deal with isolating thousands of guests even when far fewer categories are available to use.

428

Assessments

2. The two main security measures that SELinux implements on top of the virtualization layer are as follows: - Intra-guest isolation, ensuring that guests cannot attack one another, or leak information between guests - Guest/host isolation, ensuring that guests can only access and interact with the resources on the host that are needed While both are, of course, also implemented within the hypervisor code, any design flaw could lead to high-impact problems. By implementing these isolations within SELinux, we use the strength of the SELinux subsystem as an independent (and much more flexible) access control system. 3. The virt_image_t label is used for guest images when the guest is not running. Once it is running, the image is relabeled to svirt_image_t and assigned the correct set of categories. The virt_content_t label, on the other hand, is used for read-only media, such as CD images. 4. The labels can be changed by editing the guest's XML information: # virsh edit myGuestName

At the end of the XML file presented, the appropriate seclabel tags can be added to define target labels. 5. Vagrant, by default, does not have support for sVirt, but thanks to its plugin model, we can install the libvirt plugin for Vagrant. Once installed, Vagrant will use libvirt as its virtualization layer, automatically allowing us to use sVirt with Vagrant.

Chapter 10 1. SELinux works within the Linux kernel. Xen, however, is a hypervisor that sits between the hardware and the operating systems and does not use a full operating system as its base (unlike, for instance, QEMU and KVM). When we interact with Xen through Linux, we are actually interacting with Xen through the dom0 guest. Within this guest, SELinux can be running (and we even recommend it), but SELinux will remain within the virtualized guest. Xen, however, copied the SELinux approach and implemented it in its Xen Security Module framework.

Chapter 11

429

2. You can assign a label to a Xen guest by editing its configuration file (inside /etc/ xen) and adding the seclabel parameter, like so: seclabel = 'system_u:system_r:prot_domU_t'

You will need to relaunch the guest for the changes to take effect. Once the guest is booted again (using xl create), you can see its active label using xl list -Z. What are the common Xen commands that deal with XSM labels? The common commands to use are the following: - With xl list -Z, we can list the guests and their currently assigned labels. - With xl getenforce, we can query the current enforcement state of XSM. - With xl setenforce, we can set the new enforcement state of XSM. - With xl dmesg, we can see the Xen logging, including the XSM AVC log entries. - With flask-get-bool, we can query the current XSM-FLASK Booleans and their values. - With flask-set-bool, we can set a new value for an XSM-FLASK Boolean. - With flask-label-pci, we can assign a new XSM-FLASK type to a PCI device. We can also use SELinux tools that can analyze a policy file, such as seinfo or sesearch. 3. The command to load a custom policy is xl loadpolicy, or flask-loadpolicy. As long as the new policy file is not put in /boot to be automatically picked up, this loaded policy will only be active until a reboot is done, or until a new policy is loaded.

Chapter 11 1. The machinectl command does not allow administrators to change the SELinux type of the running containers. This results in all containers running by default under an unconfined domain, whereas we want confined domains to be used—preferably even with sVirt support so that containers cannot influence one another either.

430

Assessments

2. When a container is launched with a location mapping, we should use the :Z option (in case of a private mapping) or the :z option (in case of a shared mapping) to ensure that the resources are relabeled with a container-accessible SELinux type: # podman run -dit --name postgresql-test -v /srv/db/ postgresql-test:/bitnami/postgresql:Z -p 5432:5432 postgresql

Without this option, the label of the resource remains untouched, which generally means that the container runtime cannot access the resource at all. 3. We can use the udica application to generate a custom policy. The application uses the information that is provided from a podman inspect (or docker inspect) command, which shows the current container definition, and builds a custom policy specific to that container. The policy, once loaded, can then be used by the container through the --security-opt argument. 4. The main place for SELinux settings is inside the manifest under the spec configuration parameter. There, we can create the securityContext definition, which supports SELinux options through the seLinuxOptions object.

Chapter 12 1. When SELinux Booleans are changed through the /sys/fs/selinux/ booleans filesystem, the changes are not automatically committed. For that to occur, you also need to write the value 1 into /sys/fs/selinux/commit_ pending_bools. 2. The sesearch command is used to query the active policy, and can be used to query the impact of SELinux Booleans as well. Add the -b argument to limit the query to rules that are influenced by the SELinux Boolean. 3. When an SELinux policy module is loaded, it is assigned a priority that tells the system whether it should be the active module. Administrators can load new modules at a higher priority to test them out, and remove them again, without risking that no proper SELinux rules are active on the system at all. Likewise, administrators can load a policy at a lower priority, ensuring that it is not yet active, and later on remove the module at the higher priority so that the newly loaded policy becomes active. This is unlike enabling or disabling modules, which affects all priorities.

Chapter 13

431

4. The SELinux utility audit2allow transforms all SELinux-related audit events into SELinux policy code. The code can use both the legacy style (using -M) or the reference policy style (using -R -M). Regardless of the style chosen, a loadable SELinux policy module (with the suffix .pp) will be created: # grep btmp /var/log/audit/audit.log | audit2allow -R \ -M custom_staff_su_faillog

The resulting file (custom_staff_su_faillog.pp) can be loaded using semodule -i.

Chapter 13 1. The seinfo application is used to query the policy for its type content, but not for its rules. For instance, you list the types within the policy with seinfo, but you don't query what these types can do. The sesearch application, on the other hand, is used to query the rules within the policy, but does not reveal anything about the various definitions inside the policy that are not really rules (such as attribute definitions and supported classes). Hence, the main difference is that seinfo focuses on the structure of the policy, whereas sesearch focuses on the enforcements defined within the policy. 2. Reaching a domain implies domain transitions. Hence, what we are looking for is how you can transition from your current domain (say staff_t) to the target domain (say unconfined_t) and through which means—generally, this is done by executing a binary or script that triggers a type transition. Analyzing domain transitions can be done using apol (the graphical user interface), sedta, or sepolicy transition. However, the latter might not reveal the correct paths, so it is recommended to use sedta or apol for this. 3. Information flow analysis has to take up considerably more paths than a domain transition. Domain transitions are between process domains, and only a small number of actions can trigger a transition. Information flow, on the other hand, can be effected over many, many actions. Such analysis not only needs to consider read and write statements, but also file descriptor usage, socket usage, signaling, the locking and unlocking of resources, and more besides. As a result, information flow analysis uses permission maps to identify the various permissions to check, and how important (weight) a permission is for an information flow analysis session.

432

Assessments

4. Not using the tools currently at hand. The tool that compares policies, sediff, shows the differences between the policies, but is not, by itself, capable of generating SELinux policies that contain the differences between policies. Furthermore, SELinux policy modules can only add additional rules to the active policy, not remove them. Hence, even if sediff generated compatible output, it would still not be able to generate any statements that would remove existing rules from the active policy.

Chapter 14 1. An unconfined domain is still fully controlled and enforced by SELinux. It is called unconfined because such domains are granted extensive privileges by the SELinux policy. However, unlike what the name implies, they are still somewhat confined. Permissive domains, on the contrary, are not confined. SELinux will only log violations against the policy, but it will not enforce them. 2. The SELinux sandbox utility can be used to run applications in a very restricted domain. The utility will both force the application to run in a very restricted domain (sandbox_t for regular, non-graphical end user applications, or sandbox_ xserver_t for graphical applications), as well as isolate or hide access to other system resources through the use of Linux's namespaces. 3. When the init system (such as systemd) launches a daemon, it will execute a specific binary or script for it. The label of this binary or script will generally define the target domain. For instance, if the resource is labeled with bin_t, then systemd will make sure that the target service runs as unconfined_service_t. If it is labeled with postgresql_exec_t, then the target service will run with the postgresql_t type. While other permissions are also involved (such as the source context requiring transition permissions to the target), unless we are building a new policy from scratch, switching domains will be as simple as changing the label for its executable resource, and relabeling its main locations on the filesystem (such as log locations and runtime information). 4. The default skeleton application that sepolicy generate builds contains the permissive statement, which implies that the policy will run in a permissive mode. Since this means SELinux will not enforce any controls, it is very likely that the application will work flawlessly with this policy enabled.

Chapter 15

433

That is, however, not the target state, and administrators will need to remove the permissive setting from the policy and adjust as needed.

Chapter 15 1. Many Linux distributions add services and tools that fit the distribution's purpose and principles, yet which might be contradictory to what the reference policy is about. For instance, Red Hat Enterprise Linux and its derived Linux distributions will enable unconfined domains for many applications, whereas the reference policy will strive toward confinement of all applications. As a result, many Linux distributions base their policy on the reference policy, but augment and adjust it for their specific purpose. 2. The three main policy files are the following: - A type enforcement file, with the suffix .te, which contains the rules for the SELinux policy module, focusing on its owned domains. - An interface file, with the suffix .if, which exposes the interaction patterns and privileges vis-à-vis the domains and resources owned by this SELinux policy module. These interfaces are then used by other SELinux policy modules. - A context file, with the suffix .fc, which contains the file contexts for the various paths related to this SELinux policy module. Policy modules can be created with only a single file. In that case, the build system will assume that the other files are empty. 3. The use of a permission set allows policy developers to easily adjust and extend the permission set when needed, without having to change all possible SELinux policy module code entries. This is especially critical when a new permission is added by Linux (and the SELinux subsystem). For instance, suppose that the memory map system call (map) was not present yet, and is introduced later. We would need to add the map privilege to all execute calls. By using permission sets, we can add this to the appropriate permission set only. 4. Interfaces will grant privileges to domains or roles. They do not add or remove SELinux objects to the policy. Templates, on the other hand, will generate new SELinux types, roles, Booleans, or other objects. As a consequence, templates are not allowed to be called from within any Boolean-triggered block.

434

Assessments

5. The database administration role does not use userdom_admin_user_ template because it is not a system-wide administration role, but very specific to databases. userdom_admin_user_template would grant the role many more privileges than are needed to administer databases.

Chapter 16 1. The SELinux Common Intermediate Language (CIL) is not an extension to SELinux that can be easily removed. It is at the heart of SELinux policy development and support, although mainly under the hood: all SELinux policy modules that are loaded on the system are first converted into CIL before actually being loaded in memory. The CIL format is the only format used to interact with the Linux kernel and the SELinux subsystem. Because it is used as part of the SELinux user space utilities, it is not always as obvious to administrators or developers, but it is definitely a core component within SELinux. 2. No, it is not mandatory, but is recommended. The attribute is used to refer to types and roles in a modular fashion, and to ensure that these references are valid. CIL internally requires types and roles to be defined before they are used, and without using an attribute to force such declarations, the order of loading modules might result in failures. While other attributes could be declared for this purpose, or other means introduced, the use of the cil_gen_require attribute is supported through the SELinux user space utilities. As such, aligning with this practice is recommended. 3. With CIL, developers can create additional port mappings, declaring a new type and assigning it to an available port. With other SELinux language styles, this is only possible when rebuilding the entire policy rather than through the modules. Another example is to introduce constraints. SELinux constraints are not supported to be loaded in SELinux modules using other language styles. However, while constraints are a powerful construct within SELinux, they can be confusing for administrators as constraint-related failures do not result in obvious messages, and administrators who query the current policy for allow rules may notice that allow rules exist even though things still fail.

Chapter 16

435

4. CIL supports macros, which are made part of the SELinux module (and the namespace), and which can be called using the call statement from elsewhere in the SELinux policy. We can create macros as part of a module to be similar to interfaces in reference policy, while creating macros as part of the global namespace to be similar to the support macros in reference policy.

Other Books You May Enjoy If you enjoyed this book, you may be interested in these other books by Packt:

Mastering Linux Security and Hardening Donald Tevault ISBN: 978-1-83898-177-8 • Create locked-down user accounts with strong passwords • Configure firewalls with iptables, UFW, nftables, and firewalld • Protect your data with different encryption technologies • Harden the secure shell service to prevent security break-ins • Use mandatory access control to protect against system exploits • Harden kernel parameters and set up a kernel-level auditing system • Apply OpenSCAP security profiles and set up intrusion detection • Configure securely the GRUB 2 bootloader and BIOS/UEFI

438

Other Books You May Enjoy

Cybersecurity Attacks – Red Team Strategies Johann Rehberger ISBN: 978-1-83882-886-8 • Understand the risks associated with security breaches • Implement strategies for building an effective penetration testing team • Map out the homefield using knowledge graphs • Hunt credentials using indexing and other practical techniques • Gain blue team tooling insights to enhance your red team skills • Communicate results and influence decision makers with appropriate data

Leave a review - let other readers know what you think Please share your thoughts on this book with others by leaving a review on the site that you bought it from. If you purchased the book from Amazon, please leave us an honest review on this book's Amazon page. This is vital so that other potential readers can see and use your unbiased opinion to make purchasing decisions, we can understand what our customers think about our products, and our authors can see your feedback on the title that they have worked with Packt to create. It will only take a few minutes of your time, but is valuable to other potential customers, our authors, and Packt. Thank you!

Index Symbols .pp files translating, to CIL 404, 405

A access vector cache (AVC) 59 administration interface 395 AKARI 24 anonymous pipes 143 Ansible about 179 configuring 180, 181 installing 180 role, creating 181 role, testing 181 SELinux settings, using 185, 186 used, for assigning SELinux contexts to file system resources 182 used, for loading custom SELinux policies 183, 184 used, for SELinux system administration 179 working 179, 180 Ansible modules 179

Ansible playbooks 179 Ansible roles 180 Apache mod_selinux module, using 236 Apache Extension Tool 237 AppArmor 24 application-level policies about 373 creating 392 applications policies, assigning 372 running, in explicit unconfined domain 364, 365 running, in specific policy 374, 375 running, without restrictions 362 AT_SECURE 132 audit2allow about 329 used, for creating policies 328-330 used, for generating reference policy style modules 331, 332

B Bell-LaPadula model 35 bounded domain 134

440

Index

C capabilities 26 capability 24 catalog 191 category-wide policies 373 change root (chroot) 233 Chef configuring 198 development environment, preparing 200, 201 installing 198 SELinux cookbook, creating 201, 202 SELinux settings, using 204, 205 server core, configuring 199 server core, installing 199 used, for assigning SELinux contexts to file system resources 203 used, for loading custom SELinux policies 203, 204 used, for system automation 197 working 197 workstation, installing 198, 199 Chef client 198 Chef server 198 Chef, system types Chef client 198 Chef server 198 Chef workstation 198 Chef workstation 198 CIL, application policies building 410 entry point information, adding 412, 413 extending, through trial and error 414, 415

extending, with attribute assignments 412 macros, adding 416, 417 namespaces, using 410, 411 permission sets 415, 416 CIPSO label domain-specific mappings, adding 172 enabling, with NetLabel 170 enabling, with SELinux 170 CIPSO mappings configuring 171, 172 coarse-grained policies 373 Cockpit application about 228 installing 229 user logins, restricting 229, 230 code verifying, with selint 399 Common Architecture Label IPv6 Security Option (CALIPSO) 173 Common Gateway Interface (CGI) scripts 377 Common Intermediate Language (CIL) about 37, 326, 404 .pp files, translating 404, 405 syntax 405, 406 Common IP Security Option (CIPSO) 70 CONNection MARKings (CONNMARK) 150 constraints about 136, 138 learning 138, 139 container hosting environment tuning 312 container management facilitating, with machinectl 305, 306 Container Runtime Interface for OCI (CRI-O) 307

Index 441

containers SELinux domain, modifying 310 using, with SELinux 308, 309 container_t privileges toggling, with SELinux booleans 312 control groups (cgroups) 266, 302 cookbook 198 Cron services about 230 user-specific, switching between generic contexts 231, 232 custom domains creating, with udica 311, 312 customizable types 120 custom XSM policies applying 299, 300

D daemons running, in explicit unconfined domain 365, 366 database management system (DBMS) 21 Data Definition Language (DDL) 261 datagram sockets 144 data sensitivity 266 D-Bus about 224, 225 communication 224 message flows 227 service acquisition, controlling with SELinux 226, 227 D-Bus, types session-specific D-Bus 224 system-wide D-Bus 224 default distribution policy replacing 333, 334

denial information about 72 audit2why, using 75 emails, sending 74, 75 handling 77, 78 object class 67 permissions 66 permissive mode 67 Process ID (PID) 66 process name 66 SELinux action 66 source context 66 systemd-journal, interacting with 76, 77 target context 66 target device 66 target file inode number 66 target name 66 troubleshooting, with setroubleshoot 73, 74 Denial of Service (DoS) 74 discretionary 20 discretionary access control (DAC) about 20, 65 extending, with SELinux 25 domain about 30, 31, 275 marking, as permissive 368, 369 Domain 0 (dom0) 286 domain complexity about 372 application-level policies 373 category-wide policies 373 coarse-grained policies 373 fine-grained policies 372 Domain of Interpretation (DOI) 170 domain permissions about 136 querying 137, 138

442

Index

domain transition 127 domain transition analysis about 349 apol, using 350, 351 investigating 349 sedta, using 351, 352 sepolicy, using 352-354 domain transition interface 395 domain transitions 389 dominated roles 89

E eBPF maps about 155 securing 155 eBPF programs about 155 securing 155 Enterprise Linux Repository (ELRepo) 287 Executable and Linkable Format (ELF) 132 existing policies replacing 328 updating 328 explicit unconfined domain applications, running in 364, 365 daemons, running in 365, 366 eXpress Data Path (XDP) 156 extended attribute (xattr) 106 extended Berkeley Packet Filter (eBPF) about 154 assessing 154 traffic, filtering with bpfilter 156, 157 working 154, 155 extended permission rules 341

Extra Packages for Enterprise Linux (EPEL) 180

F facts 191 file contexts declaring 387, 388 Filesystem in Userspace (FUSE) 269 fine-grained definitions constraints, adding to policy 409 creating 406 port type, defining 407, 408 roles or types 406, 407 fine-grained policies 372 first-in, first-out (FIFO) 143 Flux Advanced Security Kernel (FLASK) 293 Foreign Data Wrapper (FDW) 261

G generated policies extending 375 limitations 376 sepolicy generate command 376, 377 Gentoo reference link 287 GNU C library (glibc) 132 Grand Unified Boot loader (GRUB2) 287 greatest lower bound, largest upper bound (glblub) 45 Group Identifier (GID) 217

Index 443

H Hardware Virtual Machine (HVM) 291 high-level language (HLL) 39, 404 Host Channel Adapters (HCA) 158 host intrusion detection 51 host intrusion prevention 51 HyperText Transfer Protocol (HTTP) 68

I idempotent task 176 InfiniBand 157 InfiniBand networks memory, accessing 158 protecting 158, 159 securing 157 InfiniBand partitions access controlling 161, 162 InfiniBand port 158 InfiniBand subnet managing 159, 160 information flow analysis about 354 analyzing 354 apol, using 355-357 seinfoflow, using 357 sepolicy communicate, using 357 Information Protocol Security (IPsec) 71 interface labels 148 interfaces querying, locally 400, 401 SELinux rules, exposing through 388, 389 International Obfuscated C Code Contest download link 370 Internet Key Exchange version 2 (IKEv2) 170

Internet Relay Chat (IRC) 72 internet services daemon (inetd) applications 377 IPsec setting up 169 IPv6 CALIPSO 173

J Java Virtual Machine (JVM) 265 Just-In-Time (JIT) 155

K Kernel-based Virtual Machine (KVM) 264 kernel probes (kprobes) 155 Kubernetes configuring, with SELinux support 313-315 SELinux support, leveraging 313

L labeled IPsec enabling 169 using, with SELinux 167, 168 labeled IPsec event 71 labeled networking about 163 fallback labeling, with NetLabel 163 flow, limiting on network interface 164 old-style controls, using 166, 167 peer communication, accepting from selected hosts 165 peer-to-peer flow, verifying 165 labels assigning, to packets 151, 152

444

Index

least-privilege model 26 legacy style modules building 332 libvirt categories, controlling 278, 279 configuring, for sVirt 273-275 enhancing, with SELinux 271 guest, modifying for SELinux labels 276, 277 resource labels, customizing 278 shared resources, versus dedicated resources 272 storage pool locations, modifying 279, 281 Vagrant tool, using 281 libvirt architecture assessing 272, 273 libvirt-compatible box installing 282, 283 libvirt plugin deploying 281 reference link 281 Linux Containers project 302 Linux firewalling 148 Linux kernel configuration reference link 287 Linux security 20, 21 Linux Security Module (LSM) 22-25, 55, 292 Linux system SELinux, enabling 27, 28 LoadPin 24 local CIPSO using 173 Lockdown 24 LXC 302

M machinectl container management, facilitating 305, 306 MAC_POLICY_LOAD event 69 macros about 331, 416 querying, locally 400, 401 MAC_STATUS event 70 mandatory 21 mandatory access control (MAC) 21 MCS constrained 259 MCS translation daemon (mcstransd) 90 minions 186 mod_selinux module about 236, 237 Apache SELinux sensitivity, configuring 238 domains, modifying 239, 240 end users, mapping to domains 239 using, with Apache 236 Multi-Category Security (MCS) about 36, 257, 270 access, limiting to columns based on categories 258, 259 user domain, constraining for sensitivity range manipulation 259 using 257 Multi-Level Security (MLS) about 35, 257 access, limiting to columns based on categories 258, 259 using 257 mutual exclusions (mutexes) 142

Index 445

N named file transitions 110 named pipes 143 namespaces 370 netfilter 148, 149 NetLabel used, for enabling CIPSO label 170 NetLabel event 70 Netlink sockets 144, 145 network-facing service policies constructing 392 internal SELinux rules, handling 393 network related permissions, adding 394, 395 resources, identifying 392 service interface methods, building 395, 396 nftables transitioning to 153, 154 node labels 148 No New Privilege (NNP) 134

O Open Container Initiative (OCI) 307 OpenSSH daemon about 232 Linux users, chrooting 233-235 logging in, as sysadm_r role 233

P PAM services about 96 access, prohibiting in permissive mode 98 Cockpit application 228

configuring 228 contexts, assigning 97 Cron services 230 directories, polyinstantiating 99, 100 OpenSSH daemon 232 ParaVirtualized (PV) 291 Partition Key (P_Key) 161 peer-to-peer flow verifying 166 permission groups calling 391 permission interactions 389 permissive domains 55 Personally Identifiable Information (PII) 252 Pluggable Authentication Modules (PAM) 87 podman configuring 306 selecting, over Docker 306, 308 pods SELinux contexts, setting for 315-317 policies assigning, to applications 372 comparing 358 creating, with audit2allow 328-330 generating, with sepolicy generate command 378, 379 sediff, using to compare 358 policy capabilities 133 policy macros using 390 policy module priority 326 policy modules handling 325 listing 326, 327 loading 327, 328 removing 327, 328

446

Index

structuring 384 policy repository, Arch Linux reference link 382 policy repository, CentOS reference link 382 policy repository, Debian reference link 382 policy repository, Gentoo Linux reference link 382 policy store 40 polyinstantiation 99 PostgreSQL about 244 reconfiguring, with SEPostgreSQL 244-246 SEPostgreSQL, tuning 248 test account, creating 246, 247 process communications connection contexts, listing 147 controlling 142 conversing, over UNIX domain sockets 144 Netlink sockets 145 SCTP sockets, dealing with 145, 146 shared memory, using 142 TCP sockets, dealing with 145, 146 through pipes 143 UDP sockets, dealing with 145, 146 Puppet about 191 configuring 192, 193 installing 192, 193 SELinux settings, using 196, 197 used, for assigning SELinux contexts to file system resources 194 used, for automating system management 191 used, for creating SELinux class 193, 194

used, for loading custom SELinux policies 195 used, for testing SELinux class 193, 194 working 191 Puppet Development Kit (PDK) 193 Puppet Forge 191 Puppet manifests 191 Puppet master 191 Puppet modules 191

Q QEMU copy-on-write (QCOW2) 275 Quality of Service (QoS) 74, 161

R recipes 198 Red Hat Enterprise Linux (RHEL) 73 reference policy about 382 navigating 383 reference policy style modules building 332 generating, with audit2allow 331, 332 remote command execution (RCE) 27 Remote Direct Memory Access (RDMA) 158 role-oriented interfaces 389 root privileges restricting 26 run interface 395

S SafeSetId 24 Salt execution modules 186 SaltStack about 186 configuring 187

Index 447

installing 187, 188 SELinux settings, using 190, 191 used, for assigning SELinux contexts to file system resources 189 used, for creating SELinux state 188, 189 used, for loading custom SELinux policies 190 used, for testing SELinux state 188, 189 utilizing, to configure SELinux 186 working 186 Salt State file 186 sandbox command using 370-372 sandboxed applications sandbox command, using 370-372 SELinux sandbox 370 using 369 scope of transitions environment, executing 132 limiting 131 Linux’s NO_NEW_PRIVS, using 134, 135 unconstrained transitions, disabling 132, 134 Secure Shell (SSH) 57 secure virtualization (sVirt) 267 usage, of MCS 270, 271 used, for configuring libvirt 273-275 security association database (SAD) 167 security association (SA) 167 Security-Enhanced Linux (SELinux) about 96 access, enforcing through types 31 allow rules, querying 341, 342 apol workspaces, using 348, 349 audit events 58, 59 ausearch command, using 71, 72 AVC, tuning 60, 61

browsing, with apol 345-348 configuration settings 177 configuring, with SaltStack 186 containers, using with 308, 309 context, dissecting 30, 31 database-specific object classes 250 database-specific object permissions 250, 251 default types, using 252-254 denial information, reading 65-68 domain access, granting through roles 32, 33 enabling, on Linux system 27, 28 error recovering 178 file contexts, settings 178 framework, comparing 178 idempotent task 176 information flow, controlling through sensitivity 35, 36 kernel boot parameters, using 54 labeled IPsec, using 167, 168 Linux audit, configuring 62-64 local rules, adding of domain 386, 387 local system logger, configuring 64, 65 logging 61, 62 logging and auditing 57 other type rules, querying 343 permissive or enforcing mode, switching 52, 53 policy and state management 177 policy files, using 336 policy object information, displaying 337-339 protection, disabling for single service 55, 56 resources and objects, labeling 28-30 role-related rules, querying 344 roles, limiting through users 34, 35

448

Index

sepgsql permissions 251 sepgsql-specific functions, using 257 sesearch 340, 341 single-step analysis, performing 336 state, setting 50-52 switching on and off 50 target settings and policies 176 trusted procedures, creating 255-257 type transition rules, querying 342 used, for enabling CIPSO label 170 used, for extending DAC 25 using, with systemd’s container support 302, 303 vulnerabilities, impact reducing 26, 27 Security-Enhanced PostgreSQL (SEPostgreSQL) about 244 fallback label, creating for remote sessions 260 integrating, into network 259, 260 reference link 244 troubleshooting 249 tuning, in PostgreSQL 248 used, for reconfiguring PostgreSQL 244-246 Security Identifiers (SID) 130, 339 SECurity MARKings (SECMARK) about 148, 150 implementing 150 security policy database (SPD) 167 security policy information (SPI) 170 selint code, verifying 399 reference link 399 SELinux-aware applications 56, 57 SELinux booleans container_t privileges, toggling 312 impact, inspecting 325

listing 322, 323 values, modifying 323, 324 working with 322 SELinux Booleans 53 SELinux context setting, for pods 315-317 using 304, 305 SELinux development mode 52 SELinux domain modifying, of container 310 SELINUX_ERR event 69 SELinux event types about 68 labeled IPsec event 71 MAC_CONFIG_CHANGE event 70 MAC_POLICY_LOAD event 69 MAC_STATUS event 70 NetLabel event 70 SELINUX_ERR event 69 USER_AVC event 68 SELinux file context expressions about 115 customizable types, using 120 file_contexts files, compiling 121 local modifications, exchanging 122 modifying 117, 118 recursive context operations, optimizing 119 using 116, 117 SELinux file contexts about 104 categories, placing on on files and directories 113 copying 110-112 extended attribute (xattr), backing up 114 extended attribute (xattr), restoring 114 filesystem, relabeling 123

Index 449

fixfiles tools, using 123 information, obtaining 105-107 inheriting 109 modifying 112, 113, 122 mount options, using 114, 115 moving 110-112 multilevel security, using on files 113, 114 rlpkg tools, using 123 setfiles tools, using 123 setting, at boot with tmpfiles 124, 125 setting up 108 setting, with restorecond 124 transition rules, querying 109, 110 types, interpreting 107, 108 SELinux objects declaring 384, 385 SELinux policies content, defining 45, 46 cross-user sharing, limiting 43 defining 36, 37 distinguishing between 41 distributing 36, 37 distributing, through modules 39 MLS, enabling 41 modules, bundling in policy store 40 policy versions, incrementing 44, 45 reference link 37 unconfined domains, enabling 42, 43 unknown permissions, dealing with 41, 42 writing 37, 38 SELinux policy tuning 261 SELinux policy, for CentOS URL 334 SELinux process context about 125

domain transition 127-129 implementing 126 initial context, querying 130 memory protections, tweaking 130, 131 other transitions 130 target context, verifying 129 SELinux roles about 83 access, managing through sudo 94 contexts, defining 91, 93 contexts, validating with getseuser command 93 handling 91 runcon command, using 95 switching, to system role 96 switching, with newrole 93 SELinux rules exposing, through interfaces 388-390 SELinux sandbox 370 SELinux-secured virtualization 264 SELinux support Kubernetes, configuring with 313-315 SELinux system administration Ansible, using 179 SELinux users about 35, 83 accessible domains, listing 89 categories, managing 90, 91 creating 87-89 login, customizing for services 86, 87 login, mapping 84-86 mappings, listing 83, 84 semaphore 142 sensible module names using 330, 331 sepolicy generate command about 377 used, for generating policies 378, 379

450

Index

session ID (SID) 61 session-specific D-Bus 224 single-class permission groups using 390, 391 Smack 24 socket buffers (skb) 156 Stream Control Transmission Protocol (SCTP) 146 streaming sockets 144 stub domains (stubdoms) 296 Subnet Manager (SM) 158 system automation Chef, using 197 system calls 22 systemd 210 systemd container initializing 303 systemd device management tuning 210, 211 systemd logging about 220 logs, querying 220 SELinux-related information, retrieving 220 setroubleshoot integration, using with journal 221 tuning 210, 211 systemd services device files, handling 221 files, relabeling 216, 217 implementing 212 SELinux context, setting for service 213, 214 SELinux label, setting on device node 223 SELinux, requiring for service 215, 216 socket-based activation, using 218 transient services, using 215

tuning 210, 211 udev rules, using 221, 222 unit files 212, 213 unit operation access 219 system management automating, with Puppet 191 system-wide D-Bus 224

T Tiny Code Generator (TCG) 268 TOMOYO Linux 24 transient services 215 Trusted Computer System Evaluation Criteria (TSEC) 22 type 30 type attributes about 136 defining 136 type bounds 133 type enforcement 31 type enforcement files 384 type transitions 342

U udica custom domains, creating 311, 312 unconfined domain about 362 extending 367 new application, marking to run as 363, 364 working 362, 363 UNIX domain sockets 144 unnamed pipes 143 unprivileged domains (domU) 286

Index 451

user applications adding 396, 397 USER_AVC event 68 user-based access control (UBAC) 34 User Identifier (UID) 217 user-level policies adding 398 user-managed temporary files 396 user-oriented SELinux context 82 userspace object managers 211

V Vagrant boxes configuring 283 reference link 282 Vagrant tool deploying 281 reference link 281 using, with libvirt 281 virtual file system (VFS) 143 virtualization about 264, 265 domains, reusing 267, 268 fine-tuning, with SELinux policy 268-270 risks, reviewing 266, 267 virtualization layers guest 265 host 265 hypervisor 265 image 265 virtual machine 265 virtualization, types application virtualization 265 full-system emulation 264 native virtualization 264 OS-level containerization 265

OS-level virtualization 265 paravirtualization 265 Virtual Local Area Networks (VLANs) 158 Virtual Network Computing (VNC) 275

X X Desktop Group (XDG) 396 Xen about 286 building, from source 288-290 executing, with Xen-enabled Linux kernel 287 installing 287 unprivileged guest, creating 290, 292 Xen hypervisor 286 Xen Security Modules (XSM) about 286, 292, 293 booleans, using 298 hardware resources, labeling 299 logs, querying 297 manipulating 296 policy, querying 298, 299 state, defining 296 XSM-enabled Xen executing 293 rebuilding 293, 295 XSM-FLASK 293 XSM labels rebuilding 295

Y Yama 24